2026-05-06

Final Out-of-Sample Results

Given that the original GCP1 data stream is no longer operational following the unforeseen infrastructure issue in late March/ early April 2026, the preregistered out-of-sample simulation has now effectively been completed.

As many readers know, the project has focused on a fully prospective out-of-sample evaluation of the forecasting framework originally introduced in Holmberg (2024), together with a newer regime-dependent extension based on a three-state Hidden Markov Model (HMM3).

Importantly, the purpose has never been to argue that Global Consciousness Project (GCP) data “predicts markets” in any simple unconditional sense. Rather, the study has investigated a narrower and more conditional hypothesis: namely, whether GCP-derived coherence measures may contain weak informational structure during periods characterized by elevated and structurally coherent market stress.

The figure below summarizes the relative cumulative out-of-sample performance of the benchmark and GCP-dependent forecasting models.

Below are the final out-of-sample results.

Full Out-of-Sample Results

(Including U.S. Government Shutdown)

Metric Model A (Bench) Model B (GCP) Difference

Hit Rate 67.96% 66.85% -1.10 pp

T. Return 70.90% 77.79% +6.89 pp

Sharpe R. 0.41 0.44 +0.04

Across the full evaluation period, the GCP-dependent model outperformed the benchmark model in cumulative return and Sharpe ratio, but modestly underperformed in unconditional directional accuracy. Importantly, however, the full sample contains the 2025 U.S. government shutdown period associated with budget negotiations. This period introduced unusually strong policy-driven market dynamics that affected essentially all macro-financial variables needed for prices in the econometric framework. Rapid political repricing and intervention-driven sentiment shifts were not meaningfully captured by the included explanatory variables.

For that reason, the paper also reports a shutdown-excluded specification.

Shutdown-Excluded Results

(Excluding U.S. Government Shutdown)

MetricModel A (Benchmark)Model B (GCP-dependent)Difference

Metric Model A (Bench) Model B (GCP) Difference

Hit Rate 65.16% 66.45% +1.29 pp

T. Return 52.86% 64.40% +11.54 pp

Sharpe R. 0.28 0.34 +0.06

Once the shutdown period is excluded, the GCP-dependent model outperforms the benchmark model across all three preregistered performance metrics.

This is important because the original preregistration specified hit rate, total return, and Sharpe ratio as the primary evaluation metrics before completion of the out-of-sample period.

The results therefore first look mixed but became substantially stronger once the structurally exceptional shutdown period was separated from the broader market environment.

Why Exclude the U.S. Shutdown?

The key point is not that the shutdown period should be “ignored.” The full sample remains fully disclosed and analyzed in the paper. The issue is instead that the shutdown represented a very unusual policy-driven regime in which:

• fiscal negotiations,
• political intervention risk,
• abrupt expectation shifts,
• and non-economic valuation dynamics
  dominated the market environment.

The econometric framework was designed primarily around volatility structure, cross-market transmission, and regime-dependent coherence dynamics — not around rapidly changing political bargaining processes. In other words, the shutdown period likely introduced a different information-generating process than the one the models were originally designed to evaluate.

This becomes particularly relevant once the regime decomposition is considered.

Regime Results

During the live out-of-sample evaluation, it became increasingly apparent that the contribution of the GCP-derived variables was regime dependent rather than unconditional. Consequently, an additive three-state Hidden Markov Model (HMM3) framework was introduced to classify market environments into:

• Calm,
• Noisy Stress,
• and Coherent Stress.

The results showed essentially no difference between the benchmark and GCP-dependent models during Calm or Noisy Stress periods.

The positive contribution emerged primarily during Coherent Stress environments — periods characterized by elevated but structurally persistent and synchronized market stress.

Once the shutdown period was excluded, the GCP-dependent model improved directional accuracy during Coherent Stress periods by approximately 1.72 percentage points relative to the benchmark model. This is consistent with the revised interpretation developed in the paper being written: the GCP-derived variables do not appear to function as broad unconditional forecasting variables, but may instead interact conditionally with volatility structure and systemic synchronization during periods of coherent market stress.

Final Remarks

The results therefore ends in a more cautious but also more theoretically refined position than the original study. In fact, the results do not support a strong claim that GCP data “predicts markets” in a stable unconditional sense. However, the findings do appear increasingly consistent with a weak regime-dependent framework in which collective-coherence measures may contain economically relevant information during periods characterized by elevated systemic coherence and synchronized stress dynamics.

A final complication emerged in late March/ early April 2026 when an unforeseen service issue affecting the original GCP1 infrastructure rendered the live GCP1 data stream unusable. This effectively ended the original out-of-sample simulation earlier than intended.

The natural next step is therefore not simply to continue the original GCP1-based framework, but rather to rebuild and align the forecasting structure with the newer GCP2 environment. In practice, this transition has already begun and forms the basis of the newer framework currently implemented on Coherence Analytics, although this will be expanded upon further in future work.

Importantly, the transition from GCP1 to GCP2 was not mechanical. GCP1 Max[Z] and the newer GCP2-derived coherence measures are not directly interchangeable, as they reflect different infrastructure, aggregation procedures, and potentially different statistical properties. Any future GCP2-based forecasting framework should therefore be viewed as a new prospective modeling environment rather than a seamless continuation of the original GCP1-based setup.

More detailed discussion of the results, regime structure, and the transition toward GCP2-based modeling will be presented in the final paper.

2026-05-05

Market update, data transition and regime context

By late March 2026, the original GCP1 data feed became unavailable due to unforeseen hosting-related issues. As a result, the formal out-of-sample simulation based on the original GCP1 Max[Z] metric effectively ends at that point. This is important because the main evaluation should remain based on the same data-generating process used in the preregistered model design.

After the GCP1 disruption, the analysis was extended on an exploratory basis using GCP2 data. This required constructing a derived MaxZ measure from the GCP2 coherence metric. However, this measure is not fully equivalent to the original GCP1 Max[Z] variable. The original GCP1 measure is more directly related to deviations in the mean of the RNG network, while the GCP2-derived coherence-based MaxZ appears to capture more of the variance or dispersion structure in the signal.

This means that GCP1 and GCP2 should not be treated as directly interchangeable. The two signals may become more comparable only when variance shifts and mean shifts move together. The current results suggest that this may occur during periods of Coherent Stress, when market attention becomes synchronized and collective-event structure is stronger. GCP2 is described as a newer distributed RNG-based project, and recent discussion of GCP2-based MaxZ construction also notes the use of GCP2 network coherence to generate a normalized daily z-score and extract the maximum value.

Model performance

Despite the data transition, the GCP-dependent model performed strongly during this period. In the full sample, Model B clearly outperformed the control model:

• Total return: 77.79% vs 70.90%, a gain of 6.89 percentage points
• Sharpe ratio: 0.44 vs 0.41, a gain of 0.04

This is notable because the model did not merely improve risk-adjusted performance; it also improved directional accuracy in the full sample. That makes the result stronger than the earlier interpretation, where the main claim was that Model B was “right at more important times.” Here, the GCP-dependent model was both more accurate and more profitable over the relevant full-sample period. The reduced sample remains weaker, again pointing towards the need of data to extract the signal.

Interpretation

The most plausible interpretation is that the GCP signal remains regime-dependent. During Coherent Stress, markets are more likely to move around shared narratives, such as geopolitical risk, energy prices, inflation expectations, and synchronized risk repricing. In such periods, the GCP2-derived MaxZ may capture a related coherence structure, even though it is not identical to GCP1 Max[Z].

This may explain why Model B now outperforms Model A despite the shift from GCP1 to GCP2 data. If Coherent Stress produces both mean shifts and variance shifts in the underlying RNG network, then a variance-sensitive GCP2 coherence measure may still retain predictive information relevant to the original GCP1-based framework.

Conclusion and next steps

The formal out-of-sample simulation should still be evaluated using the original GCP1 data up to the point where that data feed became unavailable. The GCP2-based extension should be treated separately, as an exploratory robustness and transition exercise.

That said, the results are interesting and promising. The fact that the GCP-dependent model outperformed the control model during a period requiring a switch to a non-identical GCP2-derived signal suggests that the underlying effect may not be tied narrowly to one specific data construction.

The next step should therefore be to clearly separate:

1. The preregistered GCP1-based OOS evaluation
2. The exploratory GCP2 transition test
3. A future formal model using GCP2 coherence-derived MaxZ.

The stronger claim is not that GCP1 and GCP2 are interchangeable. They are not. Rather, the more defensible interpretation is that, during periods of Coherent Stress, both signals may capture a related underlying structure linked to synchronized collective attention. In such environments, where market dynamics are increasingly driven by shared narratives rather than dispersed information, this structure may become relevant for forecasting.

For further context, the interested reader may explore the newly developed GCP2-based, regime-dependent model available on the Coherence Analytics platform. This model is estimated across the same cross-market inputs used in the S&P 500 framework and provides an initial view of how coherence-based signals behave under different market regimes.

:

2026-04-02

Market update and regime context

Since mid-March, markets have been dominated by developments related to the ongoing Middle East conflict and its implications for global energy supply. Oil prices remained elevated and volatile throughout the period, reflecting both disruption risks in the Strait of Hormuz and shifting expectations around potential de-escalation.

At the same time, central banks maintained a cautious stance. While inflation risks increased due to higher energy prices, policymakers largely refrained from decisive action, instead signalling a readiness to respond if needed. This created an environment where uncertainty was driven more by expectations and communication than by concrete policy moves.

Market behaviour during this period can best be described as episodic and headline-driven:

- Relief rallies emerged when oil prices temporarily eased or de-escalation appeared possible

- These were quickly reversed as tensions resurfaced or expectations shifted

- Volatility remained elevated, but not disorderly

Indeed, March as a whole saw broad equity weakness and significant drawdowns across strategies, driven by a reassessment of risk premia amid energy shocks and geopolitical uncertainty. Naturally, all this feeds in to the assessed regime…

From the HMM perspective, this period was predominantly characterised by Coherent Stress, interspersed with shorter pockets of Noisy Stress. This reflects a market where movements were not purely random, but instead aligned around shared narratives, particularly those linked to energy markets, inflation expectations, and geopolitical risk.

In practical terms, this is a regime where:

-   Cross-asset correlations increase

-   Attention becomes synchronised

-   Market moves are more “coherent” than noisy

Model performance

Within this Coherent Stress-dominated environment, the GCP-enriched model (Model B) again demonstrated relative strength. During the period from March 17 onward, Model B produced correct directional forecasts on two key dates — March 17 and March 27 — where the control model (Model A) underperformed. Both instances occurred during periods of elevated market attention and structured volatility, consistent with the conditions under which the GCP signal is hypothesised to add value.

These forecast differences, while limited in number, were impactful enough to shift cumulative performance, resulting in Model B finishing the full simulation ahead of Model A, both in terms of total return but also with regards to the models sharp ratio (that takes risk into consideration). Note however, that dispite this the hit rate was lower, possibly due that the simulation period contained a long period of market calm (i.e. a period during which much of the GCP data signal added less clear value). Note also that the reduced sample continues to show weaker performance for Model B, indicating sensitivity to sample selection and regime composition.

These results reinforce a key insight: The GCP-based signal does not improve forecasting uniformly but rather adds value selectively during periods of coherent market stress.

The decline in hit rate, despite improved total return and Sharpe ratio, is particularly informative as it suggests that Model B is not correct more often, but that it is correct at more important times, especially during periods where market moves are driven by shared attention and systemic factors.

Conversely, in calmer or less structured environments, the model tends to underperform — consistent with earlier observations.

Conclusion and next steps

The final phase of the simulation provided a meaningful out-of-sample test under a regime dominated by Coherent Stress. The fact that Model B outperformed specifically during this period — and did so through a small number of high-impact forecast improvements — is consistent with the broader hypothesis of state-dependent predictive value.

While the results remain exploratory and should be interpreted with appropriate caution, the consistency of the pattern across different segments of the sample is notable. The next step will be to formalise these findings. I will therefore proceed with writing a paper that summarises the methodology, empirical results, and interpretation of the GCP-based model in a structured and rigorous manner.

2026-03-17

Market update and regime context

Since 7 March, markets have remained in what this framework classifies as a Coherent Stress regime (visible in the figure below as the extended green segment). This corresponds to a period of elevated volatility, stronger cross-market synchronization, and a market environment driven more by shared macroeconomic and geopolitical shocks than by isolated idiosyncratic news. The dominant driver has been the escalation around Iran, which has affected global energy markets and revived inflation concerns through higher oil prices and renewed supply uncertainty. At the same time, trade-policy uncertainty has not disappeared, and political pressure on the Federal Reserve has added to questions about policy credibility and the broader macro regime. Taken together, these factors have kept markets in a more fragile and stress-sensitive state than during the calmer periods seen earlier in the sample.

Looking across the longer sample, the overall pattern remains broadly consistent.
Model A has generally performed better during calmer, lower-volatility periods, while Model B has tended to hold up relatively better when market stress becomes more synchronized and collective attention increases. The most recent period, however, adds an important nuance. Since 27 February, the two models have performed almost identically in cumulative terms. The current lead for Model A therefore does not stem from a fresh divergence during the latest stress window, but rather reflects underperformance by Model B that accumulated earlier in the sample.

This matters for interpretation as the current environment is one in which Model B should, in principle, become more informative. As such, the evidence remains mixed.

A technical point is also worth noting. The simulations here still rely on the GCP1 network, which currently operates with only around nine active devices. Such a limited network reduces signal stability and increases the likelihood that potentially meaningful effects are diluted by measurement noise. For that reason, a transition toward the GCP2 dataset is becoming increasingly important. Work on that transition is underway, and an experimental version of a state-dependent model using GCP2 data is already available on the full platform.

In summary, markets remain stress-dominated, and the latest out-of-sample simulations suggest that Model B’s performance has stabilized.

2026-03-07

Market update and regime context

Since 22 February, markets have remained in what the framework classifies as a Coherent Stress regime. In the figure above this is visible as the extended green segment, indicating a period of persistently elevated volatility and more synchronized market moves rather than the short-lived volatility spikes typically seen in calmer environments.

The news flow broadly supports this interpretation. Fiscal uncertainty in the U.S., continued trade policy tensions, and—perhaps most importantly—geopolitical developments related to Iran have kept volatility elevated relative to the earlier calm baseline. As a result, the regime classification has remained stress-dominant for most of this period.

Looking at the longer sample in the chart also highlights a pattern that has appeared repeatedly:

• Model A tends to outperform during calm (low-VIX) periods, visible in the yellow segments.
• Model B tends to perform relatively better during Coherent Stress, when collective attention and cross-market synchronization increase.

Since 22 February, markets have largely remained in the stress regime, which in principle is the environment where Model B’s structure should be most informative. However, the brief calm window at the end of February was enough to push Model A ahead of Model B once again in the cumulative simulations. This illustrates how sensitive performance can be to regime shifts and why keeping track of the prevailing market regime is important before making decisions.

One additional technical note is worth mentioning. The simulations above still rely on the GCP1 network, which currently operates with only about nine active devices. That limited network size reduces signal stability and makes an update to the GCP2 dataset increasingly necessary. Work on that transition is underway, and more information on this will follow in the near future.

For now, the key point is that markets remain in a stress-dominated regime, and further tracking will show whether the current phase develops into the deeper synchronization historically associated with clearer separation between the models.

2026-02-22

Market update and regime context

Since 8 February, the regime structure has shifted. What initially looked like episodic volatility has developed into a more persistent "Coherent Stress" regime. In the chart, the post-Feb 8 window is predominantly green, with only brief calm interruptions.

This lines up with the news flow. The U.S. has been dealing with renewed discussion on fiscal sustainability, including a partial shutdown affecting DHS from 14 February, which has been visible in risk sentiment and headline sensitivity. At the same time, policy uncertainty around tariffs remained a live theme, and markets reacted sharply when the U.S. Supreme Court struck down sweeping tariffs, supporting a relief move late in the week.

Volatility has not simply spiked and faded. VIX levels have remained elevated versus the earlier calm baseline (roughly around the low-20 area in mid-February), which is consistent with a stress-led regime classification. Cross-market co-movement has also been more synchronized during downside moves, which matches the “stress color” in the chart.

Model performance in the late February window

The two models have behaved very similarly since 8 February. The cumulative paths remain closely aligned, and no meaningful divergence has opened up during this stress phase. This differs from earlier episodes where stress clustering coincided with relative GCP outperformance.

Interpretation-wise, the current evidence suggests that persistent VIX elevation alone may not be sufficient. The stress environment is visible, but incremental model separation has not expanded materially. The relevant condition may require deeper synchronization, broader cross-asset coherence, or more extreme clustering than what has been observed so far.

In short, the regime is stress-dominant, yet performance remains largely symmetrical. Further out-of-sample tracking is still necessary before drawing firmer conclusions.

2026-02-08

Market update and regime context

Over the past two weeks, global markets have been shaped by a combination of elevated geopolitical uncertainty and episodic relief rallies. Key drivers included renewed volatility around U.S. trade and tariff rhetoric, continued sensitivity to Middle East developments, and uneven reactions to late-cycle macro data releases. This resulted in intermittent volatility spikes rather than a single, sustained stress episode, producing a market environment best described as regime mixing.

According to the three-state HMM classification, the period alternated primarily between CoherentStress and Calm, with multiple short transitions rather than a clean regime lock-in.

Model performance under mixed regimes

In this mixed regime environment, the GCP-data dependent model once again outperformed the control model. As shown in the updated simulations, the cumulative return of the GCP model pulled ahead during periods associated with CoherentStress, while maintaining parity during calmer intervals. The control model tracked markets reasonably well overall, but lagged during those windows where volatility, attention, and cross-market co-movement increased simultaneously.

Importantly, this performance differential emerged after the original out-of-sample cutoff (2025-12-31). While this means the results must be interpreted cautiously, the behaviour is consistent with earlier findings: the contribution of GCP-derived information appears to be regime-dependent, becoming more informative during structured stress rather than during uniformly calm conditions.

Taken together, the recent period reinforces a key working hypothesis of the framework: GCP-linked signals do not act as a constant predictor, but instead modulate forecast performance when markets oscillate between calm and coherent stress. The fact that the latest outperformance occurred in a regime-mixed setting is particularly notable, as it suggests sensitivity not only to outright stress, but also to transitional market phases where collective attention and uncertainty rise together.

Further out-of-sample tracking will be needed before drawing stronger conclusions, but the current update aligns well with the broader regime-conditional interpretation developed earlier.

2026-01-24

In line with the speculation outlined below, the political and financial turmoil during the week strongly favoured the GCP-dependent model such that it once again outperforms the control model. Although this occurred after the out-of-sample cutoff date (2025-12-31), the result is nonetheless highly suggestive of pronounced state dependence in how the GCP data covaries with markets.

On a related note, a new Hidden Markov state model for S&P 500 returns, using the same set of covariates as the model presented in this blog, has been estimated across all three states. This model is made publicly available through live updates on the webpage. It should be noted, however, that this is a newly estimated model and not the one presented in the present study.

2026-01-15

This chart shows cumulative out-of-sample performance (index = 100 at start) for the GCP-enhanced model versus the control model, alongside a buy-and-hold benchmark. The shaded background indicates the HMM3 regimes used in the analysis but during the OOS window, only two regimes were present—Calm and Coherent Stress. The grey band highlights the 2025 U.S. federal government shutdown period, during which the signal temporarily faded.

A key takeaway is that performance differences appear to be regime-dependent: the models behave similarly in calm periods, while divergence becomes more pronounced during elevated, more coherent stress conditions. This pattern is consistent with the hypothesis that any incremental contribution from GCP data is strongest when collective attention is unusually synchronised.

2026-01-12

Update: Regime Dependence, MaxZ (GCP), and What the Current Evidence Can Show

The continued analysis of the preregistered out-of-sample (OOS) work has now been complemented with a regime-based diagnostic using a three-state Hidden Markov Model (HMM3) applied to volatility-related features (VIX level, |ΔVIX|, and |returns|). The full sample spans 6,453 observations and separates naturally into three distinct regimes:

• Calm (low volatility and low return variability; n ≈ 2,700)
• Coherent stress (elevated volatility with more structured dynamics; n ≈ 2,900)
• Noisy stress (high volatility and disordered market dynamics; n ≈ 850)

While the precise labels are heuristic, the regimes differ materially in both volatility level and volatility dynamics, which is the key point for interpretation.

Regime-specific model building

Within each regime, I ran a forward-add specification search starting from a minimal baseline and iteratively adding predictors that increased in-sample R², while enforcing model hierarchy (i.e. retaining main effects whenever interaction terms are included). Each regime was estimated twice:

• The NO-GCP model excludings MaxZ and all MaxZ-related interactions
• The WITH-GCP model allows lagged MaxZ and a broad set of MaxZ interaction terms (with volatility measures and cross-market variables)

This design ensures that the two models are structurally comparable and differ only in the inclusion of the GCP-related information set.

Results: a clearly state-dependent contribution

The incremental contribution from MaxZ varies sharply across regimes:

Regime R² (NO-GCP) R² (WITH-GCP) ΔR²

Calm 0.211 0.215 +0.004

Stress type 1 0.262 0.270 +0.008

Stress type 2 0.465 0.491 +0.027

Two points should be highlighted .

First, the MaxZ contribution is not uniform. In calm markets it is marginal, but in stress regimes it is modest to substantially larger, adding nearly three percentage points to the models explanatory power (R²). This is an order-of-magnitude difference relative to the calm state.

Second, this pattern closely mirrors earlier findings (Holmberg 2020–2024), which suggest that any GCP-related signal is unlikely to act as a constant “always-on” predictor. Instead, it appears to become informative primarily when markets are volatile, unstable, and stressed.

Implications for power and interpretation

These diagnostics also clarify why the current OOS window is structurally underpowered to test the hypothesis on an unconditional basis. If the unconditional effect averages to only 1–3 percentage points in hit rate—because it is active mainly in specific regimes—then several thousand OOS observations (i.e. multiple years) are required for robust confirmation AHowever, a preregistration can still be informative under these conditions. Its value is not to force premature rejection or confirmation, but to lock in the hypothesised regime dependence, model comparison framework, and evaluation metrics ex ante, thereby preventing post hoc reinterpretation as more data accumulates.

What this means

These results are in-sample and exploratory by construction. Maximising R² within regimes can overfit, and the regime classification itself is model-dependent. The findings should therefore not be read as proof of a causal effect. Rather, they serve as a diagnostic map indicating where a MaxZ-related signal is most likely to matter and which interaction channels (e.g. MaxZ × volatility, MaxZ × cross-market dynamics) appear most relevant.

Next steps

As discussed below, a plausible way forward is to extend the analysis econometrically by fitting an updated, regime-aware model and evaluating its out-of-sample performance regime by regime as new data arrives over the coming quarters. After final quality and stability checks, this extension, including hypotheses, regime definitions, and decision rules, will be preregistered on the Open Science Framework (OSF). Any subsequent updates will be posted here transparently as part of that preregistered framework.

2026-01-12

The continued analysis of the out-of-sample simulation results, as well as an anaysis of the underlying data, shows that the out-of-sample window was structurally underpowered to test the hypothesised GCP data market effect on an unconditional basis. Given a plausible unconditional effect size of roughly 1–3 percentage points in hit rate—consistent with Holmberg (2020–2024) and prior GCP findings—several thousand out-of-sample observations would be required for robust confirmation (i.e., severeal years). At the same time, the results align closely with past findings that point towards a regime-dependent mechanism, with the effect largely inactive in calm markets but plausibly reaching 5–15 percentage points in hit rate during periods of structured (and coherent) volatility. The observed directional patterns found in the out.of-sample simulation are also consistent with this interpretation, providing limited but meaningful support for a small, state-dependent effect.

A way to deepen our understanding is to extend the analysis by fitting an updated, regime-aware econometric model and to examine its out-of-sample performance across different market regimes over the coming months. Updates will be posted here and incorporated as an extension to the OSF preregistration framework.

2026-01-10

This post provides a high-level overview of the preregistered out-of-sample (OOS) simulation results for the period June–December 2025, with a particular focus on how model performance varies across different market regimes. Rather than treating the OOS window as a single homogeneous period, the analysis highlights distinct volatility stages—ranging from calm conditions, through intermediate stress, to episodes of persistent and turbulent stress—that emerge naturally from a three-state Hidden Markov Model (HMM) applied to the VIX (and with implications to the results that will be discussed in forthcoming posts).

The accompanying figure illustrates how these regimes differ not only in average volatility levels but also in their distributional characteristics. Calm regimes are tightly clustered at low VIX values, coherent stress regimes occupy an intermediate but elevated volatility range, and noisy stress regimes are associated with both higher and much more dispersed VIX outcomes. Framing the OOS results in this staged manner helps clarify why predictive relationships that appear robust in some periods weaken or disappear in others.

At this stage, the analysis remains descriptive and exploratory. The aim is not to draw definitive conclusions about model superiority, but to situate the OOS results within a regime-dependent context that is consistent with earlier findings. In particular, it motivates interpreting under- or out-performance during parts of the OOS window as potentially driven by shifts between qualitatively different market environments, rather than as a simple breakdown of the underlying relationships.

2026-01-09

This post provides an initial, high-level summary of the preregistered out-of-sample (OOS) simulation results covering June–December 2025. The analysis is descriptive and intended to clarify broad patterns rather than draw final conclusions.

Overall performance

Across the full preregistered OOS window, the control model (Model A) outperformed the GCP-augmented model (Model B) on unconditional metrics such as total return, hit rate, and Sharpe ratio. While Model B showed intermittent periods of stronger performance, these were not sufficient to dominate aggregate results. The performance gap widens further in the reduced specification, indicating that Model B is more sensitive to sample size and information density.

That said, Model B performed competitively overall. This would be unlikely if the GCP component were pure noise and suggests that the preregistered mechanistic premise—nonlinear, interaction-based, and state-dependent effects—remains consistent with the data, even though the preregistered directional prediction (“Model B beats Model A overall”) is not supported in this window.

Regime dependence

Aggregate performance masks substantial regime dependence. Both threshold-based and Hidden Markov Model (HMM) diagnostics indicate that the preregistered period is overwhelmingly dominated by low-volatility conditions, with higher-volatility episodes occurring only sporadically.

Without assigning fixed labels to the detected states, the results suggest that the Max[Z] signal is not uniformly informative over time. Its contribution appears weakest during intermediate or transitional market conditions, while strengthening in more clearly defined regimes, both during particularly calm and stable periods and during brief volatility spikes (see the picture below). Because these regimes occur infrequently in the preregistered window, this dependence "brings down" the unconditional performance metrics, which helps explain why Model A ultimately performs better overall.

This interpretation is supported by late-December dynamics, where Model B recovered some relative performance during the final trading days of 2025, consistent with the regime-dependent behaviour identified by the diagnostics.

Relation to earlier findings

These results are however broadly consistent with earlier econometric work showing that GCP-related effects are nonlinear and context dependent rather than persistent across all market conditions. Previous studies found stronger GCP contributions during periods of elevated collective attention and structured market stress.

The preregistered window, however, also includes episodes of unstructured political risk, notably surrounding the U.S. government shutdown where volatility likely reflects institutional uncertainty and headline noise rather than coherent collective attention. In such environments, the GCP-data signals would not be expected to perform well, helping to explain the weaker aggregate performance observed later during the simulation period.

Next steps

With the preregistered simulation period now complete, ongoing work focuses on:

• more explicit modelling of regime dependence,
• refining regime identification,
• and applying these insights to future research using GCP-2 data, where regime-aware designs will be central from the outset.

Further updates will follow as this analysis progresses.

2026-01-03

Market Wrap-Up: December 20– December 31, 2025

Over the final trading days of 2025, global equity markets were characterized by thinner liquidity, selective risk-taking, and mild year-end consolidation, as investors balanced strong full-year performance against lingering valuation and earnings uncertainties.

In the U.S., equities closed the year with solid double-digit gains—led by technology and AI-related sectors—but sentiment softened toward year-end as profit-taking and reduced holiday liquidity weighed on prices. However, the "post-Fed-cut" environment remained supportive in principle, yet attention increasingly shifted toward the sustainability of earnings growth, capital expenditure intensity, and the longer-term payoff from large AI investments. Asian markets showed a similarly subdued tone, with muted trading activity amid shortened sessions, though full-year performance across major indices remained robust.

Simulation Update (Cumulative, Out-of-Sample)

During the final trading period of December 20–31, cumulative out-of-sample performance continued to reflect a regimea-dependent divergence between the control model (Model A) and the GCP-enhanced specification (Model B). While both models performed equally under the full dataset specification, cumulative metrics showed that Model B remained modestly behind the control, with underperformance more pronounced in the reduced sample. Hit-rate and risk-adjusted return measures both deteriorated for the GCP-enhanced model relative to the control, consistent with an environment characterized by thinner liquidity, selective risk-taking, and valuation-driven positioning rather than heightened collective stress.

Taken together, these results reinforce the interpretation that GCP-based signals contribute value primarily under conditions of elevated collective attention and systemic uncertainty, but naturally lose explanatory power in more orderly market regimes.

Nex steps

The next phase of work will focus on a detailed examination of the underlying data to better understand the drivers of the observed results. In particular, attention will be directed toward identifying whether the divergence in cumulative out-of-sample performance can be attributed to a regime shift in market conditions, as hypothesized, or whether alternative explanatory factors—such as changes in liquidity, volatility, or sectoral leadership—offer a more compelling account.

This analysis will assess to what extent the empirical outcomes align with the hypothesis specified in the preregistered framework, and whether periods of underperformance should be interpreted as theoretically consistent with a signal in more orderly, policy-driven environments, or as evidence pointing to necessary refinements of either the hyothesis or the modeling approach. Importantly, both possibilities are informative: confirmation would strengthen confidence in the conditional nature of the effect, while deviations would provide guidance on how the framework may be improved.

An intensive analysis period thus now lies ahead, during which the data will be interrogated using both descriptive and formal diagnostic tools. Insights gained from this process will directly inform future research directions, including the design and evaluation of subsequent models and extensions. In particular, the findings will feed into separate, forward-looking work using the GCP2 dataset, where lessons from the current study will be applied to refine hypotheses, model structure, and regime identification strategies.

2025-12-21

Market Wrap-Up: December 11– December 19, 2025

From 11 to 19 December, global equity markets turned more volatile as investors reassessed the durability of the recent risk rally, shifting focus from monetary easing to valuation, earnings visibility, and investment intensity.

In the U.S., sentiment softened after the Federal Reserve’s rate cut as attention moved quickly from policy support to concerns around earnings and capital expenditure, particularly in the technology sector. A mid-period repricing of AI-linked stocks, driven by growing scrutiny of investment scale and payback timelines, triggered a brief risk-off phase and raised questions about stretched expectations. Sentiment improved toward the end of the period as softer inflation narratives and renewed interest in market leaders offered some support, though risk appetite remained selective.

Simulation Update (Cumulative, Out-of-Sample)

During December 11– December 19 both models performed equally, even though conditions were mixed. The GCP-enhanced Model B cumulative out-of-sample performancecontinued to lag the control. However, data shows that the pattern of underperformance appears context-dependent.

The evidence thus remains most consistent with a regime-conditional contribution from the GCP data, where its value emerges during periods of heightened collective attention and systemic stress, but naturally recedes in more orderly, policy-driven markets. From this perspective, the current out-of-sample underperformance should be interpreted as further evidence that its relevance is conditional.

____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 38.86% , Model B (GCP): 36.39% -> GCP data effect: -2.57%

B&H benchmark: 8.11%

Hit Rate

Model A (Control): 73.21.% , Model B (GCP): 69.64% -> GCP data effect: -3.57%

B&H benchmark (# positive return days): 59.82%

Sharpe Ratio

Model A (Control): 0.48, Model B (GCP): 0.44-> GCP data effect: -0.03

B&H benchmark: 0.12

_____________________________________________________________________________

2025-12-11

Market Wrap-Up: November 28 – December 10, 2025

Global equity markets traded with a mix of resilience and caution from the final week of November through early December as investors balanced strong corporate earnings, shifting monetary policy expectations, and mixed macroeconomic data.

In the U.S., stocks finished late November with the S&P 500 and Nasdaq advancing for multiple sessions and closing November around gains, supported by optimism over a likely Federal Reserve rate cut and robust performances from key technology and megacap names. The tone carried into early December as traders awaited clearer signals on monetary policy and by 10 December, the Federal Reserve delivered a widely anticipated 25-basis-point rate cut in a closely watched decision that reflected ongoing concerns about the labor market and inflation. While the Fed signaled a more cautious stance on additional cuts in 2026, investors interpreted both the move and Chair Powell’s comments as broadly supportive for risk assets. The S&P 500 rose around 0.7%, the Dow gained roughly 1.0% (nearly 500 points), and the Nasdaq also advanced, with yields on U.S. Treasuries falling as market participants digested the easing.

Across Europe, equities hovered near stability through early December, with the both the OMXS30 and STOXX 600 broadly perfomring in line with U.S. markets as investors evaluated regional economic data and awaited more clarity on U.S. monetary policy. Continued optimism around the ECB’s likely policy path, alongside resilient industrial and consumer data in parts of the region, underpinned this relative stability. In Asia, markets showed a mixed picture. While some benchmarks rose in response to global easing expectations, other markets struggled, most notably in mainland China, where inflation concerns and weaker factory orders weighed on sentiment. Japanese equities also faced pressure as rising yields and slower household spending signaled headwinds for domestic demand.

Simulation Update (Cumulative, Out-of-Sample)

Market conditions from late November into early December created a mixed but generally supportive backdrop for risk assets, with U.S. indices grinding higher ahead of the December 10 Fed meeting and global equities holding steady. Against this environment, the behaviour of the GCP-enhanced Model B showed both temporary impairment and emerging signs of recovery.

Model B however continued to lag cumulatively through early December, but much of this deficit reflects :

• the U.S. macro blackout in October/November,
• the temporary misalignment created by Sweden’s shift to winter time while the U.S. remained on daylight saving time, and
• several sentiment reversals around month-end.

The Fed’s 25 bp rate cut on 10 December introduced an additional disturbance. Model B underperformed notably in the days surrounding the Fed decision, losing relative accuracy as markets priced in the easing path, reacted to falling Treasury yields, and rotated into megacap tech.

The evolving out-of-sample results thus suggest that the GCP-enhanced model operates within distinct market regimes, with its predictive value emerging only under specific conditions. During periods of elevated market stress, pronounced volatility shifts, and stronger cross-market synchrony—such as in the 2022–2023 window, the interaction structure of the model appears to align with the prevailing dynamics, producing a clear performance edge. In contrast, the more 2025 period exhibits a markedly different environment: moderate and less shock-driven volatility, weaker alignment across international markets, and fewer days with strong GCP data signals. In this setting, the same nonlinear interactions that previously generated an advantage cease to activate, and Model B underperforms the simpler benchmark. Taken together, it currently looks like the data and outcome are most consistent with a regime-dependent effect.

____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 36.07% , Model B (GCP): 33.55% -> GCP data effect: -2.52%

B&H benchmark: 8.93%

Hit Rate

Model A (Control): 73.33.% , Model B (GCP): 69.52% -> GCP data effect: -3.81%

B&H benchmark (# positive return days): 60.95%

Sharpe Ratio

Model A (Control): 0.48, Model B (GCP): 0.44-> GCP data effect: -0.03

B&H benchmark: 0.12

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 30.96% , Model B (GCP): 15.30%

Hit Rate (Reduced)

Model A (Control): 70.48% , Model B (GCP): 61.90%

Sharpe Ratio

Model A (Control): 0.38, Model B (GCP): 0.21

The simulations exclude the trading costs and therefore reflect only pure price movements

2025-11-27

Market Wrap-Up: Mid November to 2025-11-26

U.S. market sentiment swung sharply in late November. Initially, the optimism was driven by a "rotation" back into AI and chip stocks ahead of Nvidia’s earnings, an optimism that faded after a stronger-than-expected labour market print on 21 November which pushed yields higher and dampened hopes for near-term Fed cuts. Even Nvidia’s earnings beat could not prevent a broader tech pullback as traders reassessed stretched AI valuations. By 25–26 November, however, dovish remarks from Fed officials helped stabilise sentiment, allowing Wall Street to log a four-day winning streak into Thanksgiving. Europe followed a similar pattern as hopes grew that Washington’s data blackout would soon end, but weak corporate guidance later weighed on the STOXX 600. Sentiment improved again from 25 November as softer inflation data and supportive ECB signals revived expectations of early-2026 rate cuts. Sweden’s OMXS30 broadly tracked the continental move, pressured by the global tech wobble before stabilising as risk appetite recovered.

Simulation Update (Cumulative, Out-of-Sample), 2025-11-26

After several weeks of underperformance, which seems to be largely attributable to the U.S. government’s prolonged data blackout and the timing distortions created by the Nordic/U.S. daylight-saving mismatch, the GCP-enhanced model (Model B) appears to have regained some of its predictive momentum. Over the past week, Model B’s daily forecasts improved noticeably, once again matching the accuracy of Model A. This suggests that the GCP signal may have been more sensitive than usual to how macroeconomic uncertainty was influencing sentiment in general.

Even so, the cumulative results still reflect the earlier distortions: the control model continues to hold the lead in total return, hit rate, and Sharpe ratio. However, the recent rebound in Model B’s day-to-day performance indicates that its predictive edge may not be lost, but rather temporarily suppressed by atypical market conditions. As markets transition back into a more information-rich environment, the coming weeks will show whether Model B can re-establish its previous lead or whether the longer-term drift in hit rates will prove difficult to reverse over the remainder of the out-of-sample simulation window.

____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 32.81% , Model B (GCP): 31.75% -> GCP data effect: -1.07%

B&H benchmark: 7.76%

Hit Rate

Model A (Control): 71.88.% , Model B (GCP): 68.75% -> GCP data effect: -3.13%

B&H benchmark (# positive return days): 60.42%

Sharpe Ratio

Model A (Control): 0.47, Model B (GCP): 0.45-> GCP data effect: -0.02

B&H benchmark: 0.11

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 29.83% , Model B (GCP): 14.66%

Hit Rate (Reduced)

Model A (Control): 70.83% , Model B (GCP): 61.46%

Sharpe Ratio

Model A (Control): 0.40, Model B (GCP): 0.21

The simulations exclude the trading costs and therefore reflect only pure price movements

2025-11-19

Market Wrap-Up: mid November 2025

Sentiment remained uneven through November, though risk appetite began to stabilise as U.S. markets reopened and data flow normalised following the shutdown. European equities initially climbed on 11 November amid optimism that Washington’s data blackout would be resolved, lifting cyclicals and banks, before momentum faded as mixed earnings , and cautious corporate guidance weighed on both Swedish stocks (OMXS30) and the STOXX 600. In the U.S., early weakness tied to renewed valuation concerns around Nvidia and other AI-linked names gave way to a modest rebound as confidence returned late in the week. Meanwhile, Japan’s Nikkei continued to outperform on bank strength and a rotation into value.

The GCP-data-dependent model has also started to recover alongside these reopening dynamics, though its cumulative results still reflect the mis-timed predictions made during the shutdown period.

Simulation (cumulative, out-of-sample).

• Full sample:
• • Total return: Model A (control) 31.00% vs Model B (GCP) 29.95%; B&H 8.08%.
  • Hit rate: A 71.43% vs B 68.13% (B&H 59.34%).
  • Sharpe: A 0.48 vs B 0.47 (B&H 0.10).
• Reduced sample (2025-only):
• • Total return: A 28.36% vs B 15.34%.
  • Hit rate: A 71.43% vs B 61.54%.
  • Sharpe: A 0.41 vs B 0.24.

____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 31.00% , Model B (GCP): 29.95% -> GCP data effect: -1.05%

B&H benchmark: 6.29%

Hit Rate

Model A (Control): 71.43.% , Model B (GCP): 68.13% -> GCP data effect: -3.30%

B&H benchmark (# positive return days): 59.34%

Sharpe Ratio

Model A (Control): 0.48, Model B (GCP): 0.47-> GCP data effect: -0.02

B&H benchmark: 0.10

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 28.36% , Model B (GCP): 15.34%

Hit Rate (Reduced)

Model A (Control): 71.43% , Model B (GCP): 61.54%

Sharpe Ratio

Model A (Control): 0.41, Model B (GCP): 0.24

The simulations exclude the trading costs and therefore reflect only pure price movements

2025-11-10

Market Wrap-Up: Early November 2025

Early November trading opened under a continuing U.S. federal shutdown, prolonging the economic-data blackout. The Labor Department confirmed on Nov 7 that the monthly employment report would not be published for a second month due to the shutdown—an unprecedented handicap for investors and policymakers. Coverage through Nov 8–10 described the shutdown as still in effect and now record-long, reinforcing the sense that markets were “flying blind".

Against that backdrop, U.S. equities were choppy: brief tech-led bounces faded into range-bound trading as participants lacked the usual macro anchors. In Europe, the ECB’s Oct 30 hold kept policy steady but didn’t offset the uncertainty exported from the U.S. Asia diverged as Japan continued to benefit from AI/semis momentum while China lagged on soft activity data (consistent with the late-October PMI tone).

In summary, the ongoing data blackout is the dominant macro feature since Nov 1, keeping risk appetite fragile and price discovery hesitant.

Simulation wrap (since 1 Nov)

The out of sample simulation through early November shows the performance swing toward the control model (A) has persisted and affected all main performance metrics:

• Total return: A = 24.33% vs B = 23.33% → –1.00%
• Hit rate: A = 69.05% vs B = 65.48% → –3.57%
• Sharpe: A = 0.43 vs B = 0.42 → –0.02

As can be seen in the figure, the timing of the reversal lines up with the Oct 1 shutdown and has not reverted as of Nov 7, consistent with the idea that the information-flow break (no official U.S. data) weakens the cross-market sentiment linkages your GCP-interaction terms rely on, temporarily favoring the simpler, price-driven control model. This remains a testable hypothesis rather than a causal proof; however, the chronology is clear, and the divergence has persisted throughout the shutdown. It will be interesting to see whether the GCP-dependent models “pick up” again once normal data flow resumes after the shutdown ends.

_____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 24.33% , Model B (GCP): 23.33% -> GCP data effect: -1.00%

B&H benchmark: 8.08%

Hit Rate

Model A (Control): 69.05% , Model B (GCP): 65.48% -> GCP data effect: -3.57%

B&H benchmark (# positive return days): 60.71%

Sharpe Ratio

Model A (Control): 0.43, Model B (GCP): 0.42-> GCP data effect: -0.02

B&H benchmark: 0.15

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 21.82% , Model B (GCP): 9.47%

Hit Rate (Reduced)

Model A (Control): 69.05% , Model B (GCP): 58.33%

Sharpe Ratio

Model A (Control): 0.36, Model B (GCP): 0.17

The simulations exclude the trading costs and therefore reflect only pure price movements

2025-11-02

In response to the latest results, I have now updated the full-sample model estimates through October 31, 2025. The structure of the models remains unchanged: two otherwise identical OLS models are compared, one including Max [Z] (the GCP-based predictor) and one without it. I have also preregistered two alternative specifications that loosen the link to the Nikkei 225 slightly, due to the recent decoupling between Japanese and U.S. markets.

Key takeaways so far:

• The GCP-dependent model still explains slightly more variation in S&P 500 returns than the benchmark model (Adjusted R²: 0.341 vs. 0.325).
• As in earlier results, the standalone Max [Z] term is not significant, but several interaction terms (Max [Z] × Sweden (OMXS30), Max [Z] × HK (Hang Seng), Max [Z]×VIX etc.) remain significant, consistent with the 2023 (preprint) / 2024 (JES publication ) findings.
• This suggests that if there is a GCP effect, it likely continues to operate indirectly, i.e., through interactions with other variables rather than as a direct linear predictor. This is also consistent with the hypothesis that the effect is agnostic to direction.

To keep the preregistration valid, one version of the updated models keeps the original structure exactly as before, with only the data window extended. A second preregistered version relaxes the Nikkei 225 dependence in order to test whether the econometric relationships themselves need to be reassessed.

A parallel “fresh start” version of the models will now run alongside the original specification through the end of 2025, to determine whether any performance decline is due to fading GCP signal or model drift.

Full coefficient tables, code, and preregistration updates are available on OSF: (link), with data and models here.

2025-11-01

Market Wrap & Simulation Update: October 29–31, 2025

Market wrap.
Global markets started the week on a firmer footing after the Fed delivered a widely expected 25 bps rate cut on October 29, but sentiment quickly rotated as the central bank’s softer-than-hoped guidance pushed U.S. yields and the dollar higher, cooling risk appetite. Tech-led weakness followed on October 30, with megacaps such as Meta and Microsoft dragging the Nasdaq lower, while the ECB held rates steady and European equities traded cautiously. By October 31, markets stabilised in the U.S. as the S&P 500 closed up 0.26% on strong earnings from Amazon and Apple,. Also the Nikkei gained nearly 2% on continued AI optimism, while China lagged on weak factory data. In Europe, the STOXX 600 slipped around 0.5% and Sweden’s OMXS30 fell roughly 0.6%, pressured by exporters and softer earnings momentum.

Model performance (cumulative, out-of-sample).
It finally happened… For the first time in 92 days and 81 trades, the control model (Model A) outperformed the GCP-enhanced model (Model B) in terms of hit rate. This occurred after the control model correctly forecast the market direction on October 31, while Model B did not.

This week’s sharp intraday swings and late-session sentiment reversals, somewhat surprisingly, favoured the control model. Until now, it has usually been the GCP-driven model that anticipated such shifts, which raises the question: has its predictive edge faded? Perhaps — but a possible contributing factor this week could be the seasonal time shift in Sweden, which moved to winter time while the U.S. remained on daylight saving. That one-hour misalignment disrupts the overlap between OMXS30 and U.S. trading hours, and since Model B conditions its forecast on how Max[Z] interacts with the Swedish market, this may have temporarily weakened its predictive power.

Despite this, it should be noted that the GCP-dependent model has outperformed the control for 92 days and 81 trades, and that the full-data GCP model still leads on both cumulative return and risk-adjusted performance. However, the gradual convergence in hit rates suggests that the GCP-based edge observed earlier may in fact be fading. One needs to ask why but as stated in the original publication:

“The link between GCP data, Max[Z], and stock-market returns could be more complex than the other relationships acknowledged. As such, the GCP-dependent model may require more frequent updates, which in turn could influence results.”
In Holmberg (2024): A Novel Market Sentiment Measure

Whether the current decline in performance is temporary (due to the timing shift) or structural (due to the need for recalibration) remains an open question. Nonetheless, the preregistered model simulations will continue to be posted at regular intervals. If a recalibration is performed, it will be clearly documented, shown separately, and preregistered before being evaluated.

_____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 20.94% , Model B (GCP): 22.99% -> Gain from GCP data: +2.05%

B&H benchmark: 10.06%

Hit Rate

Model A (Control): 66.67% , Model B (GCP): 65.43% -> Gain from GCP data: -1.23%

B&H benchmark (# positive return days): 60.49%

Sharpe Ratio

Model A (Control): 0.40, Model B (GCP): 0.44-> Gain from GCP data: +0.04

B&H benchmark: 0.19

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 18.49% , Model B (GCP): 12.57%

Hit Rate (Reduced)

Model A (Control): 67.50% , Model B (GCP): 61.25%

Sharpe Ratio

Model A (Control): 0.33, Model B (GCP): 0.24

The simulations exclude the trading costs and therefore reflect only pure price movements

____________________________________________________________________________________

____________________________________________________________________________________

2025-10-29

Market Wrap & Simulation Update: October 25–28, 2025

Market wrap.
Risk appetite improved to start the week, lifted by renewed optimism over potential progress in U.S.–China trade discussions and continued strength in mega-cap technology stocks. In the U.S., equities advanced steadily, with the S&P 500 and Nasdaq both extending gains as easing Treasury yields supported valuations.

Across Europe, sentiment was cautiously constructive. The STOXX 600 rose modestly, helped by rebounds in industrials and consumer discretionary sectors following a soft previous week. In Sweden, however, the OMXS30 traded largely flat, as gains in financials and energy were offset by weakness among export-oriented manufacturers sensitive to global demand.

In Asia, markets mirrored the positive tone but with notable regional variation. Japan’s Nikkei 225 advanced as the yen stabilized after recent volatility, while Hong Kong equities gained modestly after reopening from a long weekend, supported by strength in mainland technology names. Meanwhile, Chinese mainland indices moved only slightly higher as investors weighed mixed signals on policy easing and trade rhetoric.

Model performance (cumulative, out-of-sample).
Both models correctly identified the return to positive sentiment coming out of the weekend, but neither captured the out-of-sample move on Tuesday, October 28. On that day, both Model A and Model B predicted a decline, while the S&P 500 instead advanced 0.7 %, supported by renewed buying in large-cap tech and easing Treasury yields.

Even so, cumulative performance remains clearly supportive of the GCP-enhanced framework. The full-data Model B continues to outperform, with a total return of 24.54 % versus 21.83 % for the control model and 9.28 % for the buy-and-hold benchmark. Its Sharpe ratio of 0.49, compared with 0.43 for Model A, reinforces its superior risk-adjusted results despite the occasional directional miss.

The reduced-data specification continues to lag behind, highlighting the importance of richer and higher-frequency inputs, especially when market sentiment turns quickly within a trading day.

_____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 21.83% , Model B (GCP): 24.54% -> Gain from GCP data: +2.72%

B&H benchmark: 9.28%

Hit Rate

Model A (Control): 67.95% , Model B (GCP): 67.95% -> Gain from GCP data: +0%

B&H benchmark (# positive return days): 61.54%

Sharpe Ratio

Model A (Control): 0.43, Model B (GCP): 0.49-> Gain from GCP data: +0.06

B&H benchmark: 0.21

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 19.99% , Model B (GCP): 14.00%

Hit Rate (Reduced)

Model A (Control): 69.23% , Model B (GCP): 62.82%

Sharpe Ratio

Model A (Control): 0.37, Model B (GCP): 0.22

The simulations exclude the trading costs and therefore reflect only pure price movements

____________________________________________________________________________________

____________________________________________________________________________________

2025-10-22

Market Wrap & Simulation Update: October 24, 2025

Market wrap.
Global markets ended the week on a mixed but generally constructive note. In the U.S., equities rose after inflation data came in slightly below expectations, reinforcing hopes that the Federal Reserve would proceed with another rate cut before year-end. The S&P 500 gained about 0.8 %, setting a fresh record high, while the Nasdaq Composite added roughly 1 %, lifted by strong tech earnings and a continued rebound in AI-linked stocks. Treasury yields held broadly steady, and volatility eased modestly, suggesting that markets were stabilising after earlier trade-related turbulence.

In Europe, the STOXX 600 ended the session marginally lower (-0.2 %) after early gains faded as investors booked profits following a strong week. The FTSE 100 in London, however, reached a new all-time high, supported by energy and mining shares. In Sweden, the OMXS30 traded mostly flat but held near recent highs, reflecting resilience in manufacturing and industrial exporters despite global supply-chain uncertainty. In Asia, Japan’s Nikkei 225 extended gains, supported by the weaker yen and domestic stimulus expectations.

Simulation update

Both (full data) models correctly predicted the market’s direction on October 24, sustaining their joint accuracy streak. The cumulative results now stand as follows:

Total Return / Cumulative Return

Model A (Control): 20.62% , Model B (GCP): 23.31% -> Gain from GCP data: +2.69%

B&H benchmark: 9.28%

Hit Rate

Model A (Control): 68.42% , Model B (GCP): 68.42% -> Gain from GCP data: +0%

B&H benchmark (# positive return days): 60.27%

Sharpe Ratio

Model A (Control): 0.42, Model B (GCP): 0.48-> Gain from GCP data: +0.06

B&H benchmark: 0.19

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 17.34% , Model B (GCP): 11,47%

Hit Rate (Reduced)

Model A (Control): 68.42% , Model B (GCP): 61.84%

Sharpe Ratio

Model A (Control): 0.35, Model B (GCP): 0.25

The simulations exclude the trading costs and therefore reflect only pure price movements

____________________________________________________________________________________

____________________________________________________________________________________

2025-10-22

Market Wrap & Simulation Update: October 21, 2025

Market wrap.
Global markets showed renewed strength as tensions between the U.S. and China appeared to ease. Asian equities led the way: Japan’s Nikkei 225 surged ~3.4% to a record high after Japan’s new pro-stimulus prime minister was confirmed, propelling risk sentiment higher. In the U.S., optimism grew about an impending deal between Donald Trump and Xi Jinping, and hopes that the government shutdown would soon end. The S&P 500 gained ~1.1% to close at ~6,735, while the Nasdaq Composite rose ~1.4% to ~22,990. Meanwhile, the Dow Jones Industrial Average climbed ~0.47% to 46,924.74.

In Europe, the STOXX 600 advanced about 1% as banks and industrials rebounded, while Sweden’s OMXS30 mirrored the regional gain amid steady exporter performance. Volatility slipped: the CBOE VIX dropped ~12% to ~18.23, signaling reduced market stress.

Simulation update

Both models predicted the market’s direction correctly on Monday. Notably, the GCP-enhanced model (Model B) signalled the move well before the U.S. open (and well ahead of the Stockholm close), whereas the control model (Model A) only identified the signal in the final minutes before that close.

Interpretation.
The full‐data GCP model retains its performance lead in total return and Sharpe ratio, despite the hit‐rate stalemate. The fact that Model B signalled direction earlier than Model A adds further evidence that the GCP signal may provide lead insight under certain market conditions. Meanwhile, the reduced‐data specification continues to underperform, underscoring that richer data remains critical to extract incremental forecasting value.

Total Return / Cumulative Return

Model A (Control): 18.35% , Model B (GCP): 20.99% -> Gain from GCP data: +2.64%

B&H benchmark: 8.37%

Hit Rate

Model A (Control): 67.12% , Model B (GCP): 67.12% -> Gain from GCP data: +0%

B&H benchmark (# positive return days): 60.27%

Sharpe Ratio

Model A (Control): 0.39, Model B (GCP): 0.45-> Gain from GCP data: +0.06

B&H benchmark: 0.18

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 16.04% , Model B (GCP): 10.24%

Hit Rate (Reduced)

Model A (Control): 67.12% , Model B (GCP): 60.27%

Sharpe Ratio

Model A (Control): 0.32 , Model B (GCP): 0.22

The simulations exclude the trading costs and therefore reflect only pure price movements

____________________________________________________________________________________

____________________________________________________________________________________

2025-10-21

Market Wrap & Simulation Update: October 20, 2025

Global equities started the week on a strong note as easing inflation expectations and firm corporate earnings boosted risk appetite. In the U.S., the S&P 500 advanced roughly 1.1%, with the Nasdaq 100 gaining close to 1.4%, led by semiconductor and AI-linked shares. Treasury yields steadied around 4.15% on the 10-year note, while the VIX eased to just below 18, its lowest level in two weeks, signalling improved sentiment after recent volatility.

In Europe, the STOXX 600 rose about 1.3%, supported by strength in industrials and consumer discretionary stocks. The OMXS30 in Stockholm followed higher, gaining around 0.9%, as export-oriented firms extended their rebounds. Meanwhile, in Asia, the Nikkei 225 advanced modestly and Hong Kong’s Hang Seng ended slightly higher in subdued trading as investors assessed the impact of China’s latest export-control measures.

Simulation update
Both models correctly captured the rebound, keeping the hit rate stable across specifications. The cumulative results to date are:

• Total Return: Model B (GCP) = 20.99 %, Model A (Control) = 18.35 %, Buy-and-Hold = 8.37 %.
• Hit Rate: 66.67 % for both models (vs 59.72 % for Buy-and-Hold).
• Sharpe Ratio: Model B (GCP) = 0.46 vs Model A (Control) = 0.40.

Interpretation.
The parity in hit rates suggests short-term predictive convergence between the two models, yet Model B continues to hold a meaningful lead in both cumulative return and risk-adjusted performance.

*The simulations exclude the trading costs and therefore reflect only pure price movements

Total Return / Cumulative Return

Model A (Control): 18.35% , Model B (GCP): 20.99% -> Gain from GCP data: +2.64%

B&H benchmark: 8.37%

Hit Rate

Model A (Control): 67.67% , Model B (GCP): 67.67% -> Gain from GCP data: +0%

B&H benchmark (# positive return days): 59.72%

Sharpe Ratio

Model A (Control): 0.40, Model B (GCP): 0.46-> Gain from GCP data: +0.06

B&H benchmark: 0.18

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 16.04% , Model B (GCP): 10.24%

Hit Rate (Reduced)

Model A (Control): 66.67% , Model B (GCP): 59.72%

Sharpe Ratio

Model A (Control): 0.32 , Model B (GCP): 0.22

____________________________________________________________________________________

____________________________________________________________________________________

2025-10-19

Market Wrap & Simulation Update: October 15 – October 17, 2025

Market wrap.

• Wed 10/15: U.S. stocks ended mixed to slightly lower as traders digested the FOMC minutes, which signaled that policymakers remain cautious about declaring victory over inflation. The S&P 500 slipped modestly, with defensive sectors outperforming.
• Thu 10/16: Equities declined more decisively, led by regional banks and rate-sensitive sectors. Gold hit a new record high, while Treasury yields fell to multi-month lows, reflecting a more defensive tone across markets.
• Fri 10/17: Risk appetite recovered, with the S&P 500 rising about 0.5%, pacing a late-week rebound. The VIX remained elevated compared with early October but eased from the prior spike. In Europe, the STOXX 600 finished the week little changed after mid-week weakness, while Sweden’s OMXS30 continued to edge lower.

Simulation update (cumulative, out-of-sample)

Full specification:

• Total return: Model B (GCP) 19.71% vs Model A (control) 17.10%; Buy-and-Hold 7.22%.
• Hit rate: 66.20% for both Model A and Model B (parity).
• Sharpe: 0.44 (Model B) vs 0.38 (Model A) vs 0.16 (B&H)

Reduced specification:

• Total return: 9.07% (Model B) vs 14.81% (Model A).
• Hit rate: 59.15% (Model B) vs 66.20% (Model A).
• Sharpe: 0.20 (Model B) vs 0.31 (Model A).

What changed? Both models correctly captured the Wednesday rebound but failed to pick up the shift back to uncertainty on Thursday, nor did they anticipate the return of positive sentiment on Friday. One possible explanation is the idiosyncratic nature of Thursday’s stress episode combined with the elevated VIX. As a result, the hit rates remained tied at 66.20%, even though the full-data GCP model continues to lead in cumulative return and risk-adjusted performance. As before, the reduced-data version lags behind, serving as a reminder that richer inputs remain crucial.

*The simulations exclude the trading costs and therefore reflect only pure price movements

____________________________________________________________________________________

____________________________________________________________________________________

2025-10-15

Market Wrap & Simulation Update: October 13 – October 14, 2025

Optimism early in the week gave way to renewed caution as investors digested mixed signals from the U.S.–China trade front, evolving inflation expectations, and sector-specific earnings data.

On Monday, October 13, equities rebound after the previous week’s late sell-off. This as a more measured tone from Washington and Beijing helped ease fears of a renewed trade war, while strong U.S. bank earnings supported broader sentiment. As such, the S&P 500 rose 1.6%, led by chipmakers and AI-related stocks, while the Nasdaq gained 2.2%. In Europe, the STOXX 600 finished modestly higher, supported by technology and healthcare names and the OMXS30 in Stockholm also advanced,

However, by Tuesday, October 14, sentiment deteriorated once again. The trigger came from China’s reaffirmation of tighter export controls on rare-earth materials, coupled with uncertainty around U.S. fiscal gridlock and trade policy. The S&P 500 slipped ~0.4%, and the VIX volatility index climbed while European markets followed: the STOXX 600 fell 0.5%, weighed by autos and industrials, and Sweden’s OMXS30 pulled back alongside other export-heavy indices. Asian trading also remained subdued, with Hong Kong still feeling the effects of earlier regulatory tightening and the Nikkei 225 ending flat after recent record highs.

Simulation Update

The out-of-sample simulations show that both the control (Model A) and the GCP-enhanced (Model B) correctly predicted market direction on October 13, but that Model B underperformed on October 14, possibly because it conditions its prediction on the previous day’s GCP data, which in turn had signaled a return of optimism. This brings the models’ cumulative hit rates to parity at 46 out of 68 sessions (≈ 67.65%).

This convergence in hit rates marks the first such occurrence in the current simulation cycle, and it will be interesting to observe why. If Model B retains its underlying edge, the hit rates should begin to diverge again in its favor. However, the results could also indicate the need for recalibration, particularly as the number of active RNG devices has remained steady at nine.

Despite losing its short-term predictive advantage, the full-data GCP model (Model B) continues to outperform both the control model and the benchmark on cumulative returns and risk-adjusted performance. The equalized hit rates thus represent a potential inflection point suggesting that an updated calibration or a broader GCP dataset may be required to sustain forecasting precision under evolving geopolitical and market conditions. It will be interesting to follow.

*The simulations exclude the trading costs and therefore reflect only pure price movements

_____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 17.99% , Model B (GCP): 20.62% -> Gain from GCP data: +2.63%

B&H benchmark: 7.07%

Hit Rate

Model A (Control): 67.65% , Model B (GCP): 67.65% -> Gain from GCP data: +0%

B&H benchmark (# positive return days): 58.82%

Sharpe Ratio

Model A (Control): 0.42, Model B (GCP): 0.48-> Gain from GCP data: +0.07

B&H benchmark: 0.16

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 14.47% , Model B (GCP): 9.90%

Hit Rate (Reduced)

Model A (Control): 66.18% , Model B (GCP): 60.29%

Sharpe Ratio

Model A (Control): 0.31 , Model B (GCP): 0.23

________________________________________________________________________________________________________________________________________

2025-10-11

Market Wrap & Simulation Update: October 10, 2025

Sentiment turned sharply volatile. On Friday, October 10, the U.S.–China trade narrative resurfaced with a new round of trade restrictions. Beijing’s expansion of rare-earth export controls was widely interpreted as a direct escalation in the technology and defense supply-chain conflict. The move targeted materials critical to advanced semiconductors, renewable energy systems, and military technologies. In response, President Trump announced his intention to impose 100% tariffs on all Chinese imports, cancel a planned meeting with President Xi, and implement stricter export restrictions on U.S. technology sales to China. The announcement was immediately viewed as an aggressive policy pivot, reigniting fears of a renewed U.S.–China trade war.

Markets reacted swiftly. The CBOE Volatility Index (VIX) jumped nearly five points to 21.3 — its highest reading since August — marking a sharp reversal in risk sentiment, such that the S&P 500 plunged more than 2%. In Europe, the STOXX 600 declined in tandem, led lower by banks and industrials as supply-chain disruptions loomed large. The OMXS30 in Stockholm also slid from prior highs, weighed by heavy losses in export-oriented sectors such as manufacturing and technology.

Simulation Update

The sentiment shift was picked up by a market used in the econometric models (OMXS30), such that both the GCP-data enhanced Model B and the control model (Model A) correctly predicted the market’s direction. The models’ capacity to capture sentiment shifts before the close of European trading again proved resilient, even amid a fast-moving geopolitical shock.

Cumulative results (after 68 active trading days):

• Total Return:
  Model B (GCP): 20.81 % | Model A: 17.81 % | Buy-and-Hold: 7.07 %
• Hit Rate:
  Model B: 68.66 % | Model A: 67.16 % | Buy-and-Hold: 59.70 %
• Sharpe Ratio:
  Model B: 0.49 | Model A: 0.42 | Buy-and-Hold: 0.16

Under the reduced data specification, Model B continued to underperform reaffirming how less-rich data structures struggle to isolate the GCP signal under stress-driven market regimes.

Takeaways

The events of October 10 underscore the re-emergence of trade-war dynamics as a key global market driver. China’s rare-earth export restrictions and Trump’s "tariff threats" triggered a broad movement towards "risk-off", pushing volatility higher. Yet, the actively traded models studied here successfully captured the shift, and notably, if only the modest yet significant contribution from the GCP data was isolated, the signal was found meaningful.

These results reinforce how multi-dimensional data inputs, including GCP-linked indicators, can help detect subtle sentiment shifts that precede major market moves — a capability that is becoming increasingly valuable in navigating today’s nonlinear and geopolitically charged environment.

*The simulations exclude the trading costs and therefore reflect only pure price movements

_____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 17.81% , Model B (GCP): 20.81% -> Gain from GCP data: +3.01%

B&H benchmark: 7.07%

Hit Rate

Model A (Control): 67.16% , Model B (GCP): 68.66% -> Gain from GCP data: +1.69%

B&H benchmark (# positive return days): 59.70%

Sharpe Ratio

Model A (Control): 0.42, Model B (GCP): 0.49-> Gain from GCP data: +0.08

B&H benchmark: 0.16

_____________________________________________________________________________

Total Return / Cumulative Return (reduced)

Model A (Control): 14.29% , Model B (GCP): 10.07%

Hit Rate (Reduced)

Model A (Control): 65.67% , Model B (GCP): 61.19%

Sharpe Ratio

Model A (Control): 0.31 , Model B (GCP): 0.23

____________________________________________________________________

Below, I present indexed real trading results (starting at 1K) over the most recent 90-day period, based on a strategy inspired by the findings presented here. Note that the strategy employs leveraged products to mitigate the disadvantages posed by volatile and occasionally unfavourable opening prices.

________________________________________________________________________________________________________________________________________

Updated 2025-10-10

Market Wrap: October 8 – October 9, 2025

Global markets consolidated after a strong early-week rally.

On October 8, U.S. equities extended their gains as the S&P 500 rose about 0.6 % while the Nasdaq +1.1 %. Partly, this was caused about optimism surrounding AI-related and semiconductor stocks but positive market sentiment was also held up by the Federal Reserve’s meeting minutes, which reinforced expectations for additional rate cuts.

By October 9, however, market sentiment turned more cautious as the U.S.–China trade dispute re-intensified. Beijing announced expanded export controls on several rare-earth elements vital to semiconductor and defense production, a move widely interpreted as a strategic response to U.S. technology restrictions. The announcement reignited supply-chain concerns and prompted broad profit-taking in global tech stocks. The S&P 500 thus slipped about 0.3 %,, while U.S. Treasury yields ticked higher toward 4.14 %.

Across Europe, markets mirrored the cautious tone, with the STOXX 600 ending down around 0.3 % as banking and industrial stocks weakened amid rising trade tensions and lingering worries over fiscal spending in southern member states. The OMXS30 in Stockholm moved broadly in line with its European peers, easing slightly after recent highs, though strength in energy and manufacturing helped limit losses. In Asia, performance was mixed as Japan’s Nikkei 225 edged lower after reaching record levels earlier in the week, while trading in Hong Kong remained subdued following the National Day holiday and concerns over China’s policy stance.

Simulation Update

Against this backdrop, both econometric models once again correctly predicted the one-day-ahead market direction across the active markets.
After 70 active trading days, the cumulative results are as follows:

• Total Return / Cumulative Return
  Model B (GCP) → 17.62%, Model A → 14.70%, Buy-and-Hold → 9.72%
• Hit Rate:
  Model B → 68.18%, Model A → 66.67%, Buy-and-Hold → 60.61%
• Sharpe Ratio:
  Model B → 0.50, Model A → 0.41, Buy-and-Hold → 0.27

Under the reduced-data specification, however, Model B continues to underperform confirming that smaller datasets remain less able to disentangle the GCP signal from market noise.

Takeaway

The full GCP data enhanced model continues to exhibit a robust predictive edge, maintaining direction accuracy above 68% and outperforming both benchmarks and control models even as volatility re-emerges.

*The simulations exclude the trading costs and therefore reflect only pure price movements

_____________________________________________________________________________

Total Return / Cumulative Return

Model A (Control): 14.70% , Model B (GCP): 17.62% -> Gain from GCP data: +2.93%

B&H benchmark: 10.06%

Hit Rate (Full)

Model A (Control): 65.97% , Model B (GCP): 68.18% -> Gain from GCP data: +1.52%

B&H benchmark (# positive return days): 60.61%

Sharpe Ratio (Full)

Model A (Control): 0.41, Model B (GCP): 0.50-> Gain from GCP data: +0.09

B&H benchmark: 0.27

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 11.28% , Model B (GCP): 7.17%

Hit Rate (Reduced)

Model A (Control): 65.15% , Model B (GCP): 60.61%

Sharpe Ratio

Model A (Control): 0.29 , Model B (GCP): 0.20

____________________________________________________________________________________________________________________________________________________________________________________________________________

Updated 2025-10-07

Market Wrap & Simulation Update: October 4 – October 6, 2025

Even as global markets navigated heightened volatility, sentiment remained mixed across regions. In Europe, political uncertainty and renewed concerns over fiscal sustainability in several EU member states kept investors cautious. Government bond yields in Italy and France thus edged higher amid debates over budget flexibility and increased defence spending commitments, while trading in Japan was optemistic also while Hong Kong remained muted due to the National Day holiday.

In the U.S., equities demonstrated notable resilience despite fiscal gridlock and the ongoing government shutdown. The S&P 500 gained roughly 0.4% on October 6, marking a new record high, driven by strength in technology and AI-related sectors, particularly semiconductor and cloud infrastructure firms. Investor confidence in the Federal Reserve’s rate-cut trajectory continued to underpin sentiment.

Against this backdrop, the out-of-sample simulations showed that both models correctly anticipated the one-day-ahead market direction. Since the start of the simulations, the following can be noted:

• Total return: Model B (GCP) has yielded 16.62%, outperforming Model A’s 13.72% and the Buy-and-Hold benchmark’s 9.72%.
• Hit rate: Model B achieved 67.19%, versus 65.63% for Model A and 60.94% for Buy-and-Hold.
• Sharpe ratio: Model B recorded 48.26%, ahead of Model A’s 39.07% and Buy-and-Hold’s 27.30%.

Under the reduced data specification, Model B continued to underperform relative to Model A, highlighting the importance of dataset completeness for accurate signal extraction.

Takeaway: The full-data GCP model maintained its predictive edge despite heightened political and fiscal uncertainty, correctly signaling the upside and preserving its performance lead. The continued underperformance of the reduced specification reinforces how vital comprehensive data coverage is for isolating the GCP signal from background noise.

Note also that the total return calculations are now based on only trading days when all markets used in the econometric model were open, in line with the model’s specification assumptions.

*The simulations exclude the trading costs and therefore reflect only pure price movements

Focusing only trading days when all markets used in the econometric model were open

As before

Unchanged

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 13.72% , Model B (GCP): 16.62% -> Gain from GCP data: +2.90%

B&H benchmark: 9.72%

Hit Rate (Full)

Model A (Control): 65.63% , Model B (GCP): 67.19% -> Gain from GCP data: +1.56%

B&H benchmark (# positive return days): 60.94%

Sharpe Ratio (Full)

Model A (Control): 39.07%, Model B (GCP): 48.26%-> Gain from GCP data: +9.19%

B&H benchmark: 27.30%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 10.33% , Model B (GCP): 6.26%

Hit Rate (Reduced)

Model A (Control): 64.06% , Model B (GCP): 59.38%

Sharpe Ratio

Model A (Control): 27.66% , Model B (GCP): 17.59%

________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

2025-10-04

Market Wrap & Simulation Update: October 2 – October 3, 2025

U.S. equity markets extended their rally into the first days of October, even with the federal government shut down. On October 2, the S&P 500 rose ~0.1%, as gains in technology stocks helped offset concerns from weak hiring data and the lack of key macro releases. On October 3, markets again managed a modest upside, with both the S&P 500 and Dow hitting new all-time highs, even though the Nasdaq slipped slightly.

Simulation Update: During these days, both the full-data GCP-enhanced model (Model B) and the control Model A correctly predicted the market direction resulting in the following cumulative simulation results since the onset of the simulation:

• Total Return: Model B now stands at 17.08% versus 14.17% for Model A and 9.36% for Buy-and-Hold.
• Hit Rate: Model B’s hit rate is 66.67%, compared with 65.08% for Model A and 60.32% for Buy-and-Hold.
• Sharpe Ratio: Model B registers 45.88%, outperforming Model A’s 38.08& and Buy-and-Hold’s 24.85%.

In the reduced specification, Model B’s performance remains behind: 6.64% return compared to Model A’s 9.93%, with a hit rate at 60.32% versus 63.49% for Model A.

Takeaway: Against the backdrop of a U.S. government shutdown, the full-data GCP model has again demonstrated robustness, nailing direction in both sessions and sustaining its performance edge. The reduced model continues to lag, reinforcing that richer datasets remain essential to reliably extract the GCP signal.

*The simulations exclude the trading costs and therefore reflect only pure price movements

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 14.17% , Model B (GCP): 17.08% -> Gain from GCP data: +2.91%

B&H benchmark: 9.36%

Hit Rate (Full)

Model A (Control): 65.08% , Model B (GCP): 66.67% -> Gain from GCP data: +1.59%

B&H benchmark (# positive return days): 60.32%

Sharpe Ratio (Full)

Model A (Control): 38.08%, Model B (GCP): 45.88%-> Gain from GCP data: +7.80%

B&H benchmark: 24.85%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 9.93% , Model B (GCP): 6.64%

Hit Rate (Reduced)

Model A (Control): 62.49% , Model B (GCP): 60.32%

Sharpe Ratio

Model A (Control): 26.88% , Model B (GCP): 18.88%

________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________

2025-10-02

Market Wrap & Simulation Update: September 30 – October 1, 2025

Markets traded cautiously at the turn of the quarter. On September 30, U.S. equities slipped modestly, with the S&P 500 down 0.13% and the Nasdaq off 0.26%, as investors weighed the risks of a looming government shutdown and weak labor market signals from the ADP report. The defensive tone reflected worries that fiscal gridlock could disrupt economic data releases and complicate the Federal Reserve’s policy path.

On October 1, Hong Kong markets was closed for a public holiday as invester sentiment improved in the U.S. The S&P 500 rebounded 0.37% as investors largely looked through the federal shutdown and refocused on the prospect of further Fed easing.

Simulation Results:
Both the full-data GCP-enhanced model (Model B) and the control model (Model A) correctly forecast market direction across both trading days. Over the 69 trading days studied so far (if also days when some utilized markets where closed are included), cumulative results show Model B continuing to hold a strong lead:

• Total Return: Model B has delivered 17.00%, outpacing Model A’s 14.09% and the Buy-and-Hold benchmark’s 9.28%.
• Hit Rate: Model B signaled direction correctly 65.57% of the time, compared with 63.93% for Model A and 59.02% for Buy-and-Hold.
• Sharpe Ratio: Model B’s risk-adjusted performance reached 0.464, ahead of 0.385 for Model A and 0.250 for Buy-and-Hold.

In contrast, the reduced specification lagged behind, with Model B returning 6.59% against Model A’s 9.85%, and producing a weaker Sharpe ratio (19% vs. 27%).

*The simulations exclude the trading costs and therefore reflect only pure price movements

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 14.09% , Model B (GCP): 17.00% -> Gain from GCP data: +2.81%

B&H benchmark: 9.28%

Hit Rate (Full)

Model A (Control): 63.93% , Model B (GCP): 65.57% -> Gain from GCP data: +1.64%

B&H benchmark (# positive return days): 59.02%

Sharpe Ratio (Full)

Model A (Control): 38.50%, Model B (GCP): 46.44%-> Gain from GCP data: +7.94%

B&H benchmark: 25.04%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 9.85% , Model B (GCP): 6.59%

Hit Rate (Reduced)

Model A (Control): 62.30% , Model B (GCP): 60.66%

Sharpe Ratio

Model A (Control): 27.10% , Model B (GCP): 19.03%

2025-09-30

Market Wrap & Simulation Update: September 29, 2025

U.S. markets bounced back on Monday, with the S&P 500 rising roughly 0.3%, shaking of a weak week as investors regained confidence. The rebound was driven by renewed optimism about Federal Reserve rate cuts, and declining Treasury yields. Despite persistent concerns over a possible government shutdown, markets found footing, with many participants interpreting Friday’s inflation report as supportive of further easing.

_____________________________________________________________________________

From the simulation side, both the full-data GCP-enhanced model (Model B) and the GCP-independent control model (Model A) correctly predicted the market’s direction. Over the 69 active trading days studied so far, however, Model B has clearly pulled ahead, delivering a cumulative return of 16.13% versus 13.24% for Model A and 8.47% for Buy-and-Hold. Model B also achieved the highest hit rate at 65.00%, compared with 63.33% for Model A and 58.33% for Buy-and-Hold. On risk-adjusted terms, Model B’s Sharpe ratio reached 0.45, outpacing Model A (0.37) and Buy-and-Hold (0.23). By contrast, the reduced specification using GCP data underperformed as Model B’s return was 6.15%, behind Model A’s 9.40%.

Takeaway: The full-data GCP model continues to add meaningful value in one-day-ahead forecasting, signaling market direction correctly 65% of the time and outperforming both the control and Buy-and-Hold, even in periods of heightened volatility. The persistent weakness in the specifications using the reduced data highlights the importance of rich datasets for disentangling the GCP signal from background noise.

*The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 13.24% , Model B (GCP): 16.13% -> Gain from GCP data: +2.89%

B&H benchmark: 8.47%

Hit Rate (Full)

Model A (Control): 63.33% , Model B (GCP): 65.00% -> Gain from GCP data: +1.67%

B&H benchmark (# positive return days): 58.33%

Sharpe Ratio

Model A (Control): 36.96%, Model B (GCP): 44.96%-> Gain from GCP data: +8.00%

B&H benchmark: 23.38%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 9.40% , Model B (GCP): 6.15%

Hit Rate (Reduced)

Model A (Control): 61.67% , Model B (GCP): 60.00%

Sharpe Ratio

Model A (Control): 26.39% , Model B (GCP): 18.02%

____________________________________________________________________________

2025-09-29

Market Wrap & Simulation Update: September 26, 2025

U.S. stocks rebounded strongly on Friday, breaking a three-day losing streak as the S&P 500 climbed 0.6% after key inflation data came in roughly in line with expectations. The move was fueled by renewed hopes that the Federal Reserve will continue cutting rates, as well as inflows into equity funds driven by optimism around AI and solid corporate earnings.

_____________________________________________________________________________

Against this backdrop, the GCP-enhanced model (Model B) again aligned with market direction under the full specification. Cumulatively, Model B has returned 15.82%, compared with 12.94% for Model A and 8.19% for Buy-and-Hold. The hit rate for Model B is 64.41%, ahead of Model A’s 62.71% and the benchmark’s 57.63%. On a risk-adjusted basis, Model B’s Sharpe ratio of 44.5% exceeds Model A’s 35.8% and Buy-and-Hold’s 22.8%, even though the reduced specification continues to underperform.

Takeaway: The simulations continue to show that the full-data model that is conditioned on GCP data consistently adds value to one-day-ahead forecasts.

*The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 12.94% , Model B (GCP): 15.82% -> Gain from GCP data: +2.88%

B&H benchmark: 8.19%

Hit Rate (Full)

Model A (Control): 62.71% , Model B (GCP): 64.41% -> Gain from GCP data: +1.69%

B&H benchmark (# positive return days): 56.90%

Sharpe Ratio

Model A (Control): 35.8%, Model B (GCP): 44.5%-> Gain from GCP data: +8.7%

B&H benchmark: 22.8%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 9.12% , Model B (GCP): 5.87%

Hit Rate (Reduced)

Model A (Control): 61.02% , Model B (GCP): 59.32%

Sharpe Ratio

Model A (Control): 25.8% , Model B (GCP): 17.4%

____________________________________________________________________________

2025-09-26

Market Wrap & Simulation Update: September 25, 2025

Markets came in red on Thursday, with the S&P 500 falling 0.5%, continuing a three-day skid as investors parsed stronger-than-expected economic data that could dampen expectations for aggressive rate cuts.

_____________________________________________________________________________

Both models predicted the direction such that under the full specification, the GCP-enhanced model (Model B) maintained its outperformance: 15.14% cumulative return, ahead of Model A’s 12.28% and the Buy-and-Hold benchmark’s 7.55%. Model B also posted a hit rate of 63.79%, versus 62.07% for Model A and 56.90% for Buy-and-Hold. On a risk-adjusted basis, Model B’s Sharpe ratio was 43.2%, outperforming Model A (34.5%) and Buy-and-Hold (21.4%). In contrast, the reduced specification again lagged. Model B returned 5.25%, trailing Model A (8.48%).

Takeaway: The GCP-enhanced strategy in the full model maintained its predictive edge, correctly signaling the move and outperforming both the control and Buy-and-Hold. This consistency underscores that richer GCP inputs continue to add value to predictive accuracy and risk-adjusted performance. The reduced specification, however, still struggles, suggesting that broader datasets are needed to fully extract the signal from noise.

*The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 12.28% , Model B (GCP): 15.14% -> Gain from GCP data: +2.86%

B&H benchmark: 7.55%

Hit Rate (Reduced)

Model A (Control): 62.07% , Model B (GCP): 63.79% -> Gain from GCP data: +1.72%

B&H benchmark (# positive return days): 56.90%

Sharpe Ratio

Model A (Control): 34.5%, Model B (GCP): 43.2%-> Gain from GCP data: +8.7%

B&H benchmark: 21.4%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 8.48% , Model B (GCP): 5.25%

Hit Rate (Full)

Model A (Control): 60.34% , Model B (GCP): 58.62%

Sharpe Ratio

Model A (Control): 24.4% , Model B (GCP): 15.9%

_____________________________________________________________________________

2025-09-25

Market Wrap & Simulation Update: September 24, 2025

Markets slipped again on Wednesday as caution set in following remarks from Fed Chair Jerome Powell that equity valuations looked “fairly highly valued.” Tech underperformed, dragging the S&P 500 down about 0.3% by the close. The retreat underscored that despite the strength earlier in the week, the upward momentum faces headwinds from policy uncertainty and stretched valuations.

_____________________________________________________________________________

Against this backdrop, the GCP-enhanced model (Model B) made the correct directional call under the full specification, aligning with the market downward shift. Cumulatively, Model B have returned 14.57%*, compared with Model A (11.72%*) and the Buy-and-Hold benchmark (8.09%). Its hit rate stood at 63.16%, ahead of Model A (61.40%) and the benchmark (57.89%). Risk-adjusted performance continued to favor Model B, with a Sharpe ratio of 42.1%*, versus 33.4%* for Model A and 23.3% for Buy-and-Hold.

In contrast, the reduced specification continued to show cumulative underperformance. Model B returned 4.73%, lagging both Model A (7.93%) and the Buy-and-Hold benchmark (8.09%).

Takeaway: The results reinforce that GCP inputs can add value when applied to the richer dataset, improving predictive accuracy and consistency. However, reduced datasets remain vulnerable to noise, with weaker outcomes underlining the need for fuller information to extract the signal.

*The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 11.72% , Model B (GCP): 14.57% -> Gain from GCP data: +2.85%

B&H benchmark: 8.09%

Hit Rate (Full)

Model A (Control): 61.40% , Model B (GCP): 63.16% -> Gain from GCP data: +1.75%

B&H benchmark (# positive return days): 57.89%

Sharpe Ratio

Model A (Control): 33.4%, Model B (GCP): 42.1%-> Gain from GCP data: +8.7%

B&H benchmark: 23.3%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 7.90% , Model B (GCP): 4.73%

Hit Rate (Reduced)

Model A (Control): 59.65% , Model B (GCP): 57.89%

Sharpe Ratio

Model A (Control): 23.2% , Model B (GCP): 14.5%

_____________________________________________________________________________

2025-09-24

Market Wrap & Simulation Update: September 22-23, 2025

Markets began the week on a turbulent note as geopolitical headlines over the weekend, combined with renewed trade tensions, weighed on sentiment. The S&P 500 opened lower on Monday and extended its pullback, with investors retreating to safe havens while Treasury yields eased. By Tuesday, volatility remained elevated as participants digested both the external shocks and lingering uncertainty over U.S. fiscal and monetary policy.

_____________________________________________________________________________

Neither the control model nor the GCP-enhanced model anticipated the sudden turn, underscoring the challenge of capturing unpredictable shocks through econometric inputs.

Despite the market’s shift, the GCP-enhanced model (Model B) continued to show stronger cumulative performance under the full specification. Model B delivered a total return of 14.25%, compared with 11.40% for the control model (Model A) and 8.40% for Buy-and-Hold. It also achieved a higher hit rate at 62.50% versus 60.71% for Model A and 58.93% for Buy-and-Hold, alongside superior risk-adjusted returns, with a Sharpe ratio of 41.6% compared with 32.8% for Model A and 24.5% for Buy-and-Hold.

In contrast, the reduced specification again underperformed. Model B lagged with a return of 4.43%, trailing both Model A (7.63%) and Buy-and-Hold (8.40%). Its hit rate slipped to 57.14% against 58.93% for Model A, while risk-adjusted returns weakened, with a Sharpe ratio of 13.8% compared with 22.6% for the control.

Takeaway: The results reinforce that GCP inputs can add value when applied to a richer specification, improving predictive accuracy and consistency. However, reduced datasets remain vulnerable to noise, with weaker outcomes underlining the need for fuller information to extract the signal.

The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 11.40% , Model B (GCP): 14.25% -> Gain from GCP data: +2.84%

B&H benchmark: 8.40%

Hit Rate (Full)

Model A (Control): 60.71% , Model B (GCP): 62.50% -> Gain from GCP data: +1.82%

B&H benchmark (# positive return days): 58.93%

Sharpe Ratio

Model A (Control): 32.8%, Model B (GCP): 41.6%-> Gain from GCP data: +8.8%

B&H benchmark: 24.5%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 7.63% , Model B (GCP): 4.43%

Hit Rate (Reduced)

Model A (Control): 58.93% , Model B (GCP): 57.14%

Sharpe Ratio

Model A (Control): 22.6% , Model B (GCP): 13.8%

_____________________________________________________________________________

Real-life performance over the past 90 days from a portfolio built on models with insights shared here.

_____________________________________________________________________________

2025-09-20

Market Wrap & Simulation Update: September 19, 2025

U.S. equities ended the week with the S&P 500 rising about 0.5% to a fresh record high on Friday. Gains were supported by renewed optimism over Federal Reserve policy easing, as softer inflation data earlier in the week continued to anchor expectations for a rate cut at the upcoming FOMC meeting. Technology and AI-related stocks remained at the center of market momentum, while Treasury yields edged lower, providing further support to risk assets.

_____________________________________________________________________________

The GCP-enhanced model (Model B) maintained a clear edge under the full specification. It delivered a cumulative return of 15.39%, well ahead of both the control model (12.52%) and Buy-and-Hold (8.52%). Model B also achieved the strongest hit rate at 63.64% (versus 61.82% for Model A and 58.18% for Buy-and-Hold) and produced superior risk-adjusted returns, with a Sharpe ratio of 46.9%, compared with 37.4% for Model A and 25.6% for Buy-and-Hold. By contrast, in the reduced specification, Model B underperformed: it returned 4.38%, trailing Model A (7.58%) and Buy-and-Hold (8.52%). Its hit rate slipped to 57.41% (vs. 59.26% for Model A), while its Sharpe ratio of 14.1% also lagged behind Model A (23.3%).

Takeaway: The results continue to show that GCP data add consistent value in the full model, strengthening both predictive accuracy and risk-adjusted returns. In reduced datasets, however, the signal remains harder to extract from noise, leading to weaker outcomes.

Note that the Sharpe ratios for the reduced data models presented previously contained a labeling mix-up as they showed the Buy-and-Hold Sharpe for the “without GCP” model and the “without GCP” Sharpe for the “with GCP” model. This has now been corrected.

The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 12.52% , Model B (GCP): 15.39% -> Gain from GCP data: +2.87%

B&H benchmark: 8.52%

Hit Rate (Full)

Model A (Control): 61.82% , Model B (GCP): 63.64% -> Gain from GCP data: +1.82%

B&H benchmark (# positive return days): 58.18%

Sharpe Ratio

Model A (Control): 37.4%, Model B (GCP): 46.9%-> Gain from GCP data: +9.5%

B&H benchmark: 25.6%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 7.58% , Model B (GCP): 4.83%

Hit Rate (Reduced)

Model A (Control): 59.26% , Model B (GCP): 57.41%

Sharpe Ratio

Model A (Control): 23.3% , Model B (GCP): 14.1%

_____________________________________________________________________________

Real-life performance over the past 30 days from a portfolio built on models with insights shared here.

_____________________________________________________________________________

2025-09-19

Market Wrap & Simulation Update: September 17-18, 2025

Markets moved cautiously midweek as investors digested the Federal Reserve’s decision to cut rates by 25 bps, a step widely expected after weeks of softer inflation and labor market data. While the rate cut provided reassurance that monetary policy is shifting toward support, Fed officials emphasized a data-dependent path ahead, tempering hopes for a rapid easing cycle. On Wednesday, equities slipped modestly as traders weighed the cautious tone, but by Thursday sentiment improved, led by large-cap tech and AI-related names, while stabilizing Treasury yields offered further support.

_____________________________________________________________________________

Against this backdrop, the updated simulations continue to show that GCP inputs add meaningful value when incorporated into the fully specified model. Model B delivered a cumulative total return of 14.83%, comfortably outperforming both the control model (11.97%) and the Buy-and-Hold benchmark (7.99%). The hit rate for Model B was also higher at 62.96% versus 61.11% for Model A and 57.41% for Buy-and-Hold. Risk-adjusted performance followed the same pattern, with Model B achieving a Sharpe ratio of 45.8%, well ahead of Model A (36.3%) and Buy-and-Hold (24.4%).

By contrast, the reduced data model that uses the GCP data underperformed, with Model B returning only 3.88%, compared to 8.10% for Model A and 7.99% for Buy-and-Hold. Its Sharpe ratio (25.3%) however came in slightly better(Model A, 23.4%), but the overall picture is that more data is needed to disentangle the GCP signal from noise.

Takeaway: The latest results confirm that GCP data strengthens predictive accuracy and risk-adjusted performance in the full model, while reduced datasets remain less reliable.

Footnote: The simulations exclude the trading costs and therefore reflect only pure price movements.

_____________________________________________________________________________

Total Return (Full)

Model A (Control): 11.97% , Model B (GCP): 14.83% -> Gain from GCP data: +2.86%

B&H benchmark: 7.99%

Hit Rate (Full)

Model A (Control): 61.11% , Model B (GCP): 62.96% -> Gain from GCP data: +1.85%

B&H benchmark (# positive return days): 57.41%

Sharpe Ratio

Model A (Control): 36.3%, Model B (GCP): 45.8%-> Gain from GCP data: +9.5%

B&H benchmark: 24.4%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 8.10% , Model B (GCP): 3.88%

Hit Rate (Reduced)

Model A (Control): 60.38% , Model B (GCP): 56.60%

Sharpe Ratio

Model A (Control): 23.4% , Model B (GCP): 25.3%

_____________________________________________________________________________

2025-09-17

How does a portfolio using models with insights from here perform in real life?

Some of you may have wondered whether the simulation results shared here can be applied in practice. As I’ve noted in earlier posts, the level of risk one is willing to take determines the price paid for the contract, and thereby also the balance between rewards and risks.

Since parts of the simulations have occasionally been subject to updates (for example, due to enterprise system changes), I understand that some of you may have begun to question the robustness of the results, even though both the code and econometric model are preregistered.

To help mitigate such concerns, I’m sharing below the results from a real-life portfolio that uses leveraged products informed by the same insights I’ve presented here. The performance over the past 30 days speaks for itself.

If you have any questions, feel free to reach out to me by email.

/Ulf Holmberg

Market Wrap & Simulation Update: September 16, 2025

Markets eased back on Tuesday after their record-setting run, as investors shifted focus from last week’s inflation-driven rally toward the upcoming Federal Reserve decision. With most expecting a rate cut later this month, the pause reflected a mix of profit-taking and caution rather than a shift in sentiment. Technology stocks, which had been central to the recent surge, traded sideways, while defensive sectors saw modest inflows. Bond yields held steady, underscoring that positioning now hinges on Fed guidance in the days ahead.

In the simulations, the GCP-enhanced model (Model B) continued to demonstrate clear outperformance under the full specification. Model B delivered a cumulative return of 14.39%, compared with 11.54% for Model A and 7.58% for Buy-and-Hold. It also posted the highest hit rate at 63.46% (versus 61.54% for Model A and 57.69% for Buy-and-Hold) and showed stronger risk-adjusted performance, with a Sharpe ratio of 45.5%, outperforming both Model A (35.8%) and Buy-and-Hold (23.7%).

By contrast, the reduced specification underperformed, with Model B returning only 3.99%, well below Model A (8.20%) and Buy-and-Hold (7.58%). Its hit rate was lower (57.69%), compared with 61.54% for Model A, though its Sharpe ratio of 25.9% still edged ahead of Model A’s 24.0%, indicating a relative advantage in risk-adjusted terms despite weaker absolute returns.

Takeaway: The results continue to highlight that GCP data adds clear value when applied in a fully specified model, enhancing both predictive accuracy and consistency. However, in reduced datasets, the signal remains difficult to disentangle from noise, leading to weaker outcomes. Importantly, additional work has now been completed to ensure that all dates with no trades in any of the markets used for the econometric model have been excluded across all simulations—covering Model A, Model B, and the Buy-and-Hold benchmark—for both the full specification and the reduced dataset. This refinement strengthens comparability and consistency across strategies.

Footnote: The simulations exclude the trading costs and therefore reflect only pure price movements.

Total Return (Full)

Model A (Control): 11.54% , Model B (GCP): 14.39% -> Gain from GCP data: +2.85%

B&H benchmark: 7.58%

Hit Rate (Full)

Model A (Control): 61.54% , Model B (GCP): 63.46% -> Gain from GCP data: +1.92%

B&H benchmark (# positive return days): 57.69%

Sharpe Ratio

Model A (Control): 35.8%, Model B (GCP): 45.5%-> Gain from GCP data: +9.7%

B&H benchmark: 23.7%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 8.20% , Model B (GCP): 3.99%

Hit Rate (Reduced)

Model A (Control): 61.54% , Model B (GCP): 57.69%

Sharpe Ratio

Model A (Control): 24.0% , Model B (GCP): 25.9%

_____________________________________________________________________________

2025-09-16

Market Wrap & Simulation Update: September 13–15, 2025

Markets extended their record-setting run through mid-September as optimism surrounding the Federal Reserve’s upcoming policy meeting kept investor sentiment high. Softer labor data from the prior week maintained expectations for a near-term rate cut, while technology and AI-related stocks once again drove the broader rally. On Monday, the S&P 500 closed at yet another record high, with Alphabet surpassing a $3 trillion market capitalization and Tesla gaining after Elon Musk disclosed a $1 billion share purchase. U.S.-China relations also provided a modest boost after officials signaled progress on a framework for TikTok’s U.S. operations. Although Nvidia slipped on renewed regulatory scrutiny in China, sector strength in semiconductors, cloud computing, and AI-related names helped sustain market momentum. Bond yields also fell, further supporting equities, as investors largely looked past inflation concerns in anticipation of Fed easing.

Against this backdrop, the updated simulations show that the GCP-enhanced model (Model B) continued to outperform Model A under the full specification. Model B delivered a cumulative return of 14.24%, compared with 11.40% for Model A and 7.72% for Buy-and-Hold. It also achieved the highest hit rate at 62.75%, versus 60.78% for Model A and 58.82% for Buy-and-Hold. Risk-adjusted performance was particularly noteworthy, with Model B posting a Sharpe ratio of 45.5%, well above both Model A (35.7%) and Buy-and-Hold (24.4%).

Takeaway: The latest results reinforce that GCP inputs add meaningful value, boosting returns, hit rates, and risk-adjusted performance beyond both the control and Buy-and-Hold strategies.

Footnote: The simulations exclude the trading costs and therefore reflect only pure price movements.

Total Return (Full)

Model A (Control): 11.40% , Model B (GCP): 14.24% -> Gain from GCP data: +2.84%

B&H benchmark: 7.72%

Hit Rate (Full)

Model A (Control): 60.78% , Model B (GCP): 62.75% -> Gain from GCP data: +1.96%

B&H benchmark (# positive return days): 58.82%

Sharpe Ratio

Model A (Control): 35.7%, Model B (GCP): 45.5%-> Gain from GCP data: +9.8%

B&H benchmark: 24.4%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 8.34% , Model B (GCP): 3.85%

Hit Rate (Reduced)

Model A (Control): 62.75% , Model B (GCP): 56.86%

Sharpe Ratio (Reduced)

Model A (Control): 24.7% , Model B (GCP): 26.6%

_____________________________________________________________________________

2025-09-13

Market Wrap & Simulation Update: September 12, 2025

Markets ended the week on a softer footing, with the S&P 500 dipping slightly after consumer sentiment data showed a four-month low and inflation expectations inched higher. While the Nasdaq managed another record close thanks to tech resilience, the broader tone turned more cautious as investors weighed the Fed’s likely rate cut against signs of strain in household outlooks. Treasury yields stabilized after Thursday’s sharp drop, and trading volumes thinned heading into the weekend.

In the simulations, the GCP-enhanced model (Model B) missed yesterday’s market shift but still maintained a strong cumulative edge under the full specification. Model B has so far delivered a total return of 13.71%, compared with 10.88% for Model A and 7.22% for Buy-and-Hold. Its hit rate remained high at 62.75%, ahead of Model A (60.78%) and Buy-and-Hold (55.36%).

Takeaway: While Model B missed yesterday’s directional move, the broader evidence continues to show that GCP data adds meaningful value in the full model.

Footnote: The simulations exclude the trading costs and therefore reflect only pure price movements.

Total Return (Full)

Model A (Control): 10.89% , Model B (GCP): 13.71% -> Gain from GCP data: +2.83%

B&H benchmark: 7.22%

Hit Rate (Full)

Model A (Control): 60.78% , Model B (GCP): 62.75% -> Gain from GCP data: +1.96%

B&H benchmark (# positive return days): 55.36%

Sharpe Ratio

Model A (Control): 34.6%, Model B (GCP): 4.44%-> Gain from GCP data: +9.8%

B&H benchmark: 23.1%

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 8.10% , Model B (GCP): 3.62%

Hit Rate (Reduced)

Model A (Control): 62.75% , Model B (GCP): 58.86%

Sharpe Ratio (Reduced)

Model A (Control): 22.7% , Model B (GCP): 26.0%

_____________________________________________________________________________

2025-09-12

Market Wrap & Simulation Update: September 11, 2025

Markets extended their record-setting run on Thursday after August inflation data came in softer than expected, reinforcing investor confidence that the Federal Reserve will cut rates later this month. The S&P 500 and Nasdaq both closed at new highs, supported by renewed strength in technology and AI-related stocks, while falling Treasury yields provided an additional tailwind for equities. The rally highlighted how central bank expectations and sector-specific momentum remain the primary drivers of market sentiment.

Against this backdrop, the GCP-enhanced model (Model B) continued to demonstrate clear outperformance under the full specification. It delivered a cumulative return of 13.76%, compared with 10.82% for Model A and 7.27% for Buy-and-Hold. Model B also posted the highest hit rate at 64.00% (versus 60.00% for Model A and 56.36% for Buy-and-Hold) and achieved better risk-adjusted performance, with a Sharpe ratio of 0.45, compared with 0.35 for Model A and 0.24 for Buy-and-Hold.

In contrast, the reduced specification underperformed: Model B returned only 3.67%, well below Model A (8.15%) and Buy-and-Hold (7.27%). Its hit rate declined to 58.00% compared with 64.00% for Model A, though its Sharpe ratio of 0.26 edged out Model A’s 0.23, indicating stronger risk-adjusted performance despite weaker absolute returns.

Takeaway: The results continue to confirm that GCP data adds meaningful value in the full model, enhancing both accuracy and risk-adjusted returns. However, in reduced datasets the signal becomes harder to disentangle from noise, and performance remains weaker.

Footnote: The simulations exclude the trading costs and therefore reflect only pure price movements.

Total Return (Full)

Model A (Control): 10.82% , Model B (GCP): 13.67% -> Gain from GCP data: +2.94%

B&H benchmark: 7.27%

Hit Rate (Full)

Model A (Control): 60.00% , Model B (GCP): 64.00% -> Gain from GCP data: +4.00%

B&H benchmark (# positive return days): 56.36%

Sharpe Ratio

Model A (Control): 0.35 , Model B (GCP): 0.45 -> Gain from GCP data: +0.10

B&H benchmark: 0.24

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 8.15% , Model B (GCP): 3.67%

Hit Rate (Reduced)

Model A (Control): 64.00% , Model B (GCP): 58.00%

Sharpe Ratio (Reduced)

Model A (Control): 0.23 , Model B (GCP): 0.24

_____________________________________________________________________________

2025-09-11

Market Wrap & Simulation Update: September 6 – September 10, 2025

After a strong start to the month, markets continued to build momentum through mid-September. Weaker labor data at the end of the prior week kept expectations for a Federal Reserve rate cut firmly in place, and by Tuesday, the S&P 500 was holding near record highs as investors positioned ahead of key inflation data. On Wednesday, the release of softer-than-expected August inflation figures sparked a renewed rally, pushing both the S&P 500 and Nasdaq to fresh records. Gains were led by technology and AI-related stocks, with Oracle surging on strong earnings, while falling Treasury yields added further support to risk assets. The rally underscored how central policy expectations and sector-specific strength remain the primary drivers of equity sentiment.

Against this backdrop, the GCP-enhanced model (Model B) also continued to show a strong edge under the full specification. It delivered a cumulative return of 12.80%, compared with 9.89% for Model A and 6.37% for Buy-and-Hold. Model B further achieved the highest hit rate at 63.27% (versus 59.18% for Model A and 6.37% for Buy-and-Hold) and posted superior risk-adjusted returns, with a Sharpe ratio of 0.43, ahead of both Model A (0.33) and Buy-and-Hold (0.21).

Footnote: All simulations exclude trading costs and therefore reflect only pure price movements.

Total Return (Full)

Model A (Control): 9.89% , Model B (GCP): 12.80% -> Gain from GCP data: +2.91%

B&H benchmark: 6.37%

Hit Rate (Full)

Model A (Control): 59.18% , Model B (GCP): 63.27% -> Gain from GCP data: +4.08%

B&H benchmark (# positive return days): 55.60%

Sharpe Ratio

Model A (Control): 0.33 , Model B (GCP): 0.43 -> Gain from GCP data: +0.10

B&H benchmark: 0.21

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 7.24% , Model B (GCP): 2.80%

Hit Rate (Reduced)

Model A (Control): 63.27% , Model B (GCP): 57.14%

Sharpe Ratio (Reduced)

Model A (Control): 0.21 , Model B (GCP): 0.24

_____________________________________________________________________________

2025-09-09

Market Wrap: September 5 – September 8, 2025

The first full week of September saw markets extend their rally as softer-than-expected labor data reinforced expectations of a Federal Reserve rate cut later this month. The S&P 500 and Nasdaq notched fresh record highs on Friday, led by strong gains in semiconductor stocks, while bond yields fell sharply as investors repositioned for easier central bank policy. By Tuesday, confidence in a near5 term Fed rate cut remained the dominant driver, keeping the S&P 500 near all-time highs as investors balanced optimism over policy easing with caution ahead of key inflation data due later in the week.

Simulation update

As trading costs now are excluded, the results reflect only pure price movements, making the strategies directly comparable.

The GCP-enhanced model (Model B) again showed a strong edge under the full specification, delivering a total return of 12.17% compared with 9.27% for Model A and 5.77% for Buy-and-Hold. Model B also achieved the highest hit rate at 61.70% (versus 57.45% for Model A and 5.77% for Buy-and-Hold) and produced better risk-adjusted returns, evident from a Sharpe ratio of 0.42, compared with 0.31 for Model A and 0.20 for Buy-and-Hold. By contrast, the GCP-dependent model using the smaller dataset for the econometric specification underperformed its GCP-independent counterpart, suggesting that more data is needed to disentangle the covariation between GCP the GCP data and market dynamics beyond what simple market indicators can capture.

Takeaway: With costs set aside, the evidence continues to show that GCP data adds clear value in the full model, improving both predictive accuracy and risk-adjusted performance. In reduced datasets, however, the signal becomes harder to separate from noise, leading to weaker outcomes.

Total Return (Full)

Model A (Control): 9.27% , Model B (GCP): 12.17% -> Gain from GCP data: +2.90%

B&H benchmark: 5.77%

Hit Rate (Full)

Model A (Control): 57.45% , Model B (GCP): 61.70% -> Gain from GCP data: +4.26%

B&H benchmark (# positive return days): 53.85%

Sharpe Ratio

Model A (Control): 0.31 , Model B (GCP): 0.42 -> Gain from GCP data: +0.11

B&H benchmark: 0.20

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 6.64% , Model B (GCP): 2.22%

Hit Rate (Reduced)

Model A (Control): 61.70% , Model B (GCP): 55.32%

Sharpe Ratio (Reduced)

Model A (Control): 0.19 , Model B (GCP): 0.23

_____________________________________________________________________________

2025-09-05

Simulation Updates Resuming

Hey, just a quick note. The file got lost in an enterprise update, but I’ve now managed to recover it. While going through things again, I looked at how the contract price moved during the day and came to the conclusion that until I have proper intraday data (which I’ll use when writing the paper), the best approach is to ignore those variations so we can focus on comparing the two strategies more clearly.

Sorry I wasn’t able to update the simulations this past week, I’ll aim to provide at least two updates a week going forward.

Market Wrap: August 30 – September 4, 2025

After Nvidia-driven gains carried the S&P 500 to fresh highs at the end of August, the new month began on a more volatile note. Early September brought a sharp pullback as a court ruling clouded the outlook for Trump-era tariffs, sparking renewed trade uncertainty and sending the S&P lower alongside rising bond yields and a flight into gold. The dip proved short-lived, however, as tech stocks stabilized midweek, while weaker manufacturing and labor market data reinforced expectations that the Federal Reserve would move ahead with a September rate cut. By Thursday, easing job figures and cooling economic momentum fueled a strong rebound, pushing the S&P 500 to another record high.

Simulation update

With trading costs excluded, the simulations now reflect only pure price movements. Under this setup, the GCP-enhanced model (Model B) strongly outperformed in the full specification, delivering 12.29% versus 9.39% for Model A and 5.88% for Buy-and-Hold, alongside higher hit rates and stronger risk-adjusted returns. By contrast, in the reduced specification, Model B lagged behind, underscoring that richer datasets are needed to fully harness the GCP signal.

Total Return (Full)

Model A (Control): 9.39% , Model B (GCP): 12.29% -> Gain from GCP data: +2.90%

B&H benchmark: 5.88%

Hit Rate (Full)

Model A (Control): 57.78% , Model B (GCP): 62.22% -> Gain from GCP data: +4.44%

B&H benchmark (# positive return days): 56.25%

Sharpe Ratio

Model A (Control): 0.33 , Model B (GCP): 0.41 -> Gain from GCP data: +0.08

B&H benchmark: 0.21

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 6.75% , Model B (GCP): 2.33%

Hit Rate (Reduced)

Model A (Control): 62.22% , Model B (GCP): 55.56%

Sharpe Ratio (Reduced)

Model A (Control): 0.24 , Model B (GCP): 0.11

_____________________________________________________________________________

2025-08-30

Market Wrap, August 26-29, 2025

Markets surged midweek as Nvidia-driven optimism pushed the S&P 500 to a fresh all-time high on Wednesday, with the Nasdaq and Dow also advancing, underscoring how dependent the broader rally had become on the performance of a single stock. Sentiment moderated on Thursday as valuations in the tech sector came under renewed scrutiny, prompting a partial rotation into defensive stocks. By Friday, the mood had turned more cautious as global equity fund inflows slowed to their weakest pace in weeks on worries over Fed independence.

Simulation update

The movemenys driven by single stocks,, was not at full recognized by the econometric models. But, nonetheless , the GCP-enhanced model (Model B) continued to outperform the control (Model A) under the full specification. Model B achieved a total return of 5.00%, compared with 4.53% for Model A, a gain of +0.47 percentage points. While this was slightly below the Buy-and-Hold benchmark (5.20%), Model B posted the highest hit rate at 60.87%, compared with 58.70% for Model A and 46.81% for Buy-and-Hold. Risk-adjusted performance was also stronger, with a Sharpe ratio of 0.22, versus 0.20 for both Model A and Buy-and-Hold, underscoring improved consistency.

By contrast, in the reduced specification, GCP data detracted from performance. Model B’s total return was 1.75%, well below Model A (5.54%) and Buy-and-Hold (5.20%).

Takeaway: The results continue to confirm that GCP inputs add value when applied in a fully specified model, enhancing both predictive accuracy and risk-adjusted returns. In contrast, under the reduced specification, the signal remains harder to separate from noise and tends to weigh on performance. Notably, Model A in the reduced setting outperformed both the market benchmark and the full-specification models, underscoring how market correlations are often fluid and subject to frequent change. At the same time, the fact that incorporating GCP data can lift outcomes by nearly half a percentage point highlights untapped potential—suggesting that if the reduced-data setting could be structured more effectively, meaningful improvements may be achievable there as well.

Total Return (Full)

Model A (Control): 4.53% , Model B (GCP): 5.00% -> Gain from GCP data: +0.47%

B&H benchmark: 5.20%

Hit Rate (Full)

Model A (Control): 58.70% , Model B (GCP): 60.87% -> Gain from GCP data: +2.17%

B&H benchmark (# positive return days): 46.81%

Sharpe Ratio

Model A (Control): 0.20 , Model B (GCP): 0.22 -> Gain from GCP data: +0.02

B&H benchmark: 0.20

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 5.54% , Model B (GCP): 1.75%

Hit Rate (Reduced)

Model A (Control): 63.04% , Model B (GCP): 54.35%

Sharpe Ratio (Reduced)

Model A (Control): 0.24 , Model B (GCP): 0.08

2025-08-26

Market Wrap, August 25, 2025

Markets softened early in the week after a strong end to the previous session. The S&P 500 slipped 0.4%, the Dow declined 0.8%, and the Nasdaq eased 0.2% as investors shifted focus to upcoming data and earnings reports. Sentiment weighed on the U.S. dollar, which stabilized after last week's slide as expectations for a September rate cut climbed, current pricing signals an 84–86% probability of a 25-basis-point cut.

Simulation update

Under the full specification, the GCP-enhanced model (Model B) continued to outperform the control (Model A). Model B achieved a total return of 5.63%, compared with 5.16% for Model A, a gain of +0.47 percentage points, and it also outperformed the Buy-and-Hold benchmark (4.86%). The hit rate for Model B was the highest at 64.29%, versus 61.90% for Model A and 48.84% for Buy-and-Hold. Risk-adjusted performance also improved, with a Sharpe ratio of 0.27, compared to 0.24 for Model A and 0.20 for Buy-and-Hold.

Total Return (Full)

Model A (Control): 5.16% , Model B (GCP): 5.63% -> Gain from GCP data: +0.47%

B&H benchmark: 4.86%

Hit Rate (Full)

Model A (Control): 61.90% , Model B (GCP): 64.29% -> Gain from GCP data: +2.38%

B&H benchmark (# positive return days): 48.84%

Sharpe Ratio

Model A (Control): 0.24 , Model B (GCP): 0.27 -> Gain from GCP data: +0.02

B&H benchmark: 0.20

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 5.59% , Model B (GCP): 2.37%

Hit Rate (Full)

Model A (Control): 64.29% , Model B (GCP): 57.14%

Sharpe Ratio

Model A (Control): 0.27 , Model B (GCP): 0.11

2025-08-23

Market Wrap, August 22, 2025

Markets rallied strongly as Federal Reserve Chair Jerome Powell’s remarks at the Jackson Hole Symposium signaled a possible pivot toward interest rate cuts. The Dow soared, marking its first record closing high of 2025, driven by renewed confidence in a September U.S. policy rate easing. The S&P 500, also gained 1.52%, reversing a five-day losing streak.

Simulation update

Under the full specification, the GCP-enhanced model (Model B) continued to outperform the control (Model A). Model B delivered a total return of 5.34%, compared with 4.86% for Model A, a gain of +0.47 percentage points. It also slightly outpaced the Buy-and-Hold benchmark (5.31%). The hit rate for Model B was the highest at 63.41%, versus 60.98% for Model A and 47.62% for Buy-and-Hold. Risk-adjusted performance also improved, with a short term Sharpe ratio of 0.26 compared to 0.23 for Model A and 0.22 for Buy-and-Hold.

Total Return (Full)

Model A (Control): 4.86% , Model B (GCP): 5.34% -> Gain from GCP data: +0.47%

B&H benchmark: 5.31%

Hit Rate (Full)

Model A (Control): 60.98% , Model B (GCP): 63.41% -> Gain from GCP data: +2.44%

B&H benchmark (# positive return days): 47.62%

Sharpe Ratio

Model A (Control): 0.23 , Model B (GCP): 0.26 -> Gain from GCP data: +0.02

B&H benchmark: 0.22

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 5.29% , Model B (GCP): 2.66%

Hit Rate (Reduced)

Model A (Control): 63.41% , Model B (GCP): 58.54%

Sharpe Ratio (Reduced)

Model A (Control): 0.25 , Model B (GCP): 0.13

2025-08-22

Market Wrap, August 21, 2025

Markets weakened on Thursday, August 21, extending their recent downtrend as sticky economic data and rising Treasury yields eroded hopes for an imminent Federal Reserve rate cut. The S&P 500 fell 0.4%, its fifth consecutive loss, while the Dow slipped 0.3% and the Nasdaq declined 0.7%. Stronger-than-expected U.S. manufacturing data pushed yields higher and strengthened the dollar, adding further pressure on equities. Sentiment was also hit by a 4.5% drop in Walmart shares after earnings disappointed, despite the retailer raising its full-year guidance. With investors now awaiting Fed Chair Powell’s remarks at the Jackson Hole symposium, expectations are shifting toward a more cautious monetary stance, leaving markets braced for continued volatility.

Simulation update

Under the full specification, the GCP-enhanced model (Model B) outperformed the control (Model A) across all metrics. Model B delivered a total return of 3.99%, ahead of Model A’s 3.53% (+0.47 percentage points), and also outpaced the Buy-and-Hold benchmark (3.73%). The hit rate for Model B was the highest at 62.50%, compared with 60.00% for Model A and 46.34% for Buy-and-Hold. Risk-adjusted performance was modestly stronger as well, with a Sharpe ratio of 0.21 versus 0.19 for Model A and 0.17 for Buy-and-Hold.

By contrast, under the reduced specification, GCP data detracted from performance. Model B’s return fell to 1.35%, significantly below Model A (3.95%) and the Buy-and-Hold benchmark (3.73%). The hit rate also declined to 57.50%, compared with 62.50% for Model A, while the Sharpe ratio dropped sharply to 0.07, versus 0.21 for Model A and 0.17 for Buy-and-Hold.

Takeaway: The results highlight that GCP data adds meaningful value when incorporated into a fully specified model, boosting both return and risk-adjusted accuracy. However, when fewer explanatory variables are used, the GCP signal appears harder to separate from noise, leading to weaker outcomes

Total Return (Full)

Model A (Control): 3.53% , Model B (GCP): 3.99% -> Gain from GCP data: +0.47%

B&H benchmark: 3.73%

Hit Rate (Full)

Model A (Control): 60.00% , Model B (GCP): 62.50% -> Gain from GCP data: +2.50%

B&H benchmark (# positive return days): 46.34%

Sharpe Ratio

Model A (Control): 0.19 , Model B (GCP): 0.21 -> Gain from GCP data: +0.02

B&H benchmark: 0.17

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 3.95% , Model B (GCP): 1.35%

Hit Rate (Reduced)

Model A (Control): 62.50% , Model B (GCP): 57.50%

Sharpe Ratio (Reduced)

Model A (Control): 0.21 , Model B (GCP): 0.07

2025-08-21

Market Wrap, August 19–20, 2025

Equities faced turbulence at the start of the week as investor sentiment weakened following a weekend dominated by geopolitical unease and monetary policy anticipation. Concerns about escalating trade frictions and heightened scrutiny on U.S. fiscal sustainability weighed on risk appetite, while attention turned toward the upcoming Jackson Hole symposium, where Fed Chair Powell is expected to clarify the central bank’s policy path.

On Tuesday, August 19, U.S. markets sold off sharply, led by technology and AI-related names. The Nasdaq fell about 1.5% and the S&P 500 with 0.6%, reflecting valuation concerns in high-growth sectors alongside broader caution ahead of Powell’s speech. The cautious tone persisted into Wednesday, August 20, with the Nasdaq down another 0.7% and the S&P 500 easing 0.2%, while the Dow managed a modest gain. Rotation into lower-valuation sectors provided some support, but sentiment remained fragile amid mixed retail earnings. Overall, the market seems to be in a "wait-and-see mode", with investors trimming exposure to riskier assets as both geopolitics and monetary policy cloud the outlook.

Simulation update

Under the full specification, the GCP-enhanced model (Model B) continues to show incremental improvements relative to the control (Model A). Model B achieved a total return of 3.82%, outperforming Model A’s 3.35% by +0.46 percentage points. While the Buy-and-Hold benchmark (4.15%) delivered the strongest overall return, Model B posted the highest hit rate (61.54%), surpassing both the control (58.97%) and benchmark (45.00%). The Sharpe ratio was also modestly higher for Model B (0.20 vs. 0.18 for the control and 0.19 for Buy-and-Hold), underscoring improved risk-adjusted performance.

By contrast, under the reduced specification, GCP integration weighed on results. Model B produced a 1.01% return, lagging both the control (3.60%) and Buy-and-Hold. These weaker outcomes suggest that when fewer data points are included, the GCP signal becomes harder to distinguish from noise.

Takeaway: The findings currently suggest that the GCP inputs add value under the full specification, improving consistency and risk-adjusted returns. However, the reduced specification highlights the need for a larger dataset to disentangle GCP-driven impct on the used market sentiment measures.

Total Return (Full)

Model A (Control): 3.35% , Model B (GCP): 3.82% -> Gain from GCP data: +0.46%

B&H benchmark: 4.15%

Hit Rate (Full)

Model A (Control): 58.97% , Model B (GCP): 61.54% -> Gain from GCP data: +2.56%

B&H benchmark (# positive return days): 45.00%

Sharpe Ratio

Model A (Control): 0.17 , Model B (GCP): 0.20 -> Gain from GCP data: +0.03

B&H benchmark: 0.19

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 3.60% , Model B (GCP): 1.01%

Hit Rate (Reduced)

Model A (Control): 60.53% , Model B (GCP): 55.26%

Sharpe Ratio (Reduced)

Model A (Control): 0.20 , Model B (GCP): 0.06

2025-08-19

Market Wrap, August 15–18, 2025

Markets ended last week with the S&P 500 and Nasdaq gaining nearly 1% and the Dow advancing around 1.7% for the week, supported by one of the strongest earnings seasons on record. On Monday, August 18, equities steadied near record highs, though the S&P 500 dipped slightly (–0.01%). Investor focus remains on upcoming retailer earnings and Fed Chair Powell’s Jackson Hole remarks later this week.

Simulation update

Under the full specification, the GCP-data-enhanced model (Model B) continues to outperform the control (Model A), delivering a higher total return (+0.47 percentage points) supported by a stronger hit rate (+2.70 percentage points). While the Buy-and-Hold benchmark achieved the highest overall return (5.02%), it lagged behind Model B on risk-adjusted performance.

Total Return (Full)

Model A (Control): 4.09% , Model B (GCP): 4.56% -> Gain from GCP data: +0.47%

B&H benchmark: 5.02%

Hit Rate (Full)

Model A (Control): 62.16% , Model B (GCP): 64.86% -> Gain from GCP data: +2.70%

B&H benchmark (# positive return days): 47.37%

Sharpe Ratio

Model A (Control): 0.23 , Model B (GCP): 0.26 -> Gain from GCP data: +0.03

B&H benchmark: 0.24

_____________________________________________________________________________

Total Return (Reduced)

Model A (Control): 4.23% , Model B (GCP): 1.62%

Hit Rate (Reduced)

Model A (Control): 61.11% , Model B (GCP): 55.56%

Sharpe Ratio (Reduced)

Model A (Control): 0.24 , Model B (GCP): 0.09

2025-08-15

Market Wrap, August 13–14, 2025

On Wednesday, markets extended their rally as investors positioned for a September rate cut by the Federal Reserve. The S&P 500 climbed about 0.3% to close at a new high, while the Dow surged over 1%, and the Nasdaq added 0.1%. On Thursday, makets where shakier as the producer Price Index unexpectedly rose 0.9% in July, stoking inflation concerns and rattling rate-cut bets. Despite the headwind, the S&P 500 managed a cautious gainer (+0.03%), while the Dow and Nasdaq held steady.

Simulation update

Under the full specification, the GCP-data-enhanced model (Model B) continues to outperform the control model (Model A) in total return (+0.47 percentage points), hit rate (+2.86 percentage points), and Sharpe ratio (+0.03). While the Buy-and-Hold benchmark achieved the highest raw return (5.33%), it lagged behind Model B in both hit rate and risk-adjusted performance.

In the reduced specification, Model B underperformed relative to Model A, with lower returns, hit rate, and Sharpe ratio—suggesting that the value of GCP data is more evident when the model is fully specified with a richer dataset.

Total Return (Full)

Model A (Control): 4.60% , Model B (GCP): 5.07%

B&H benchmark: 5.33%

Hit Rate (Full)

Model A (Control): 62.86% , Model B (GCP): 65.71%

B&H benchmark (# positive return days): 47.22%

Sharpe Ratio

Model A (Control): 0.27 , Model B (GCP): 0.30

B&H benchmark: 0.27

Total Return (Reduced)

Model A (Control): 3.79% , Model B (GCP): 1.19%

Hit Rate (Reduced)

Model A (Control): 60.00% , Model B (GCP): 54.29%

Sharpe Ratio (Reduced)

Model A (Control): 0.22 , Model B (GCP): 0.07

2025-08-13

Markets rallied sharply after July’s inflation data came in below expectations, fueling renewed optimism that the Federal Reserve will cut interest rates in September. The S&P 500 closed at a record high, up 1.1%, also supported by the extension of the U.S.–China tariff truce.

Model Performance Overview

Under the full specification, the GCP-data-enhanced model (Model B) continued to outperform the control model (Model A) across all key metrics. Model B delivered a total return of 4.77%, compared to 4.30% for Model A, alongside a higher hit rate (63.64% vs. 60.61%) and a stronger Sharpe ratio (0.2904 vs. 0.2607). While the Buy-and-Hold (B&H) benchmark posted the highest raw return at 4.96%, it lagged significantly in directional accuracy (44.12%) and slightly in risk-adjusted performance (Sharpe ratio 0.2575).

In the reduced specification, however, Model B underperformed. Its total return was 0.90% compared to 3.49% for Model A, with a lower hit rate (51.52% vs. 57.58%) and a substantially weaker Sharpe ratio (0.0555 vs. 0.2104). This suggests that the way market sentiment co-varies with the GCP data is more complex, requiring a richer dataset to disentangle subtle nuances. Consequently, in the reduced model, the inclusion of GCP data can be seen as only introducing noise, leading to a deterioration in performance consistent with what would be expected from adding irrelevant or weakly correlated variables.

Overall, the results indicate that GCP data can add meaningful value in models with richer datasets, enhancing both accuracy and risk-adjusted returns, but is less effective when the dataset is restricted.

Specification | Metric | Model A (Control) | Model B (GCP) | B&H |

-------------------------------------------------------------------------------------------------------

Full | Total Return | 4.30% | 4.77% | 4.96% |

| Hit Rate | 60.61% | 63.64% | 44.12% |

| Sharpe Ratio | 26.07% | 29.04% | 0.2575 |

Reduced | Total Return | 3.49% | 0.90% | - |

| Hit Rate | 57.58% | 51.52% | - |

| Sharpe Ratio | 0.2104 | 0.0555 | - |

------------------------------------------------------------------------------------------------------

2025-08-12

Markets retreated yesterday as investors turned cautious ahead of the U.S. CPI inflation report and digested renewed trade tensions, including news that major U.S. semiconductor companies will remit a portion of their China-related revenue to the U.S. government. These developments, combined with broader geopolitical uncertainty, pressured equities, with the S&P 500 ending the day slightly lower.

The GCP-data-enhanced model (Model B) nevertheless continued to perform and outperformd both the control model (Model A), under the full specification. Model B delivered a total return of 3.94%, compared with 3.48% for Model A , while also improving the hit rate from 59.38% (Model A) to 62.50%. Its Sharpe ratio rose to 0.2522 versus 0.2217 for Model A, indicating stronger risk-adjusted returns.

Specification | Metric | Model A (Control) | Model B (GCP) | B&H |

----------------------------------------------------------------------------------------------------------------

Full | Total Return | 3.48% | 3.94% | 3.78% |

| Hit Rate | 59.38% | 62.50% | 42.42% |

| Sharpe Ratio | 0.2217 | 0.2522 | 0.2108 |

Reduced | Total Return | 2.67% | 0.10% | - |

| Hit Rate | 56.25% | 50.00% | - |

| Sharpe Ratio | 0.1699 | 0.0088 | - |

----------------------------------------------------------------------------------------------------------------

Note: Comparing the full specification of Model A and Model B is sufficient to assess the value of GCP data, as the only difference between them is the inclusion of the GCP variable, making any performance gap directly attributable to it.

2025-08-09

Markets rebounded on the back of strong corporate earnings and optimism around potential Federal Reserve interest-rate cuts. The GCP-data-enhanced model (Model B) also continued to outperform the control model (Model A). Under the full specification, Model B achieved a 3.68% return versus 3.21% for Model A, while improving the hit rate from 58.06% (Model A) to 61.29%.

Full Specification

• Total Return: Model B (GCP-enhanced) delivered 3.68%, outpacing Model A (control) at 3.21%.
• Hit Rate: Model B correctly predicted market direction 61.29% of the time, compared to 58.06% for Model A.

Buy-and-Hold (B&H) Benchmark

• The passive strategy has delivered a total return of 4.05%, with a directional accuracy of 40.63%.

Important Notes Regarding the Simulations

• Minor coding errors were identified in both Model B using the full specification and Model A for the reduced specification. As a result, the GCP data-enhanced model began outperforming the control model in the full specification from the beginning of July. In contrast, the reduced control model seems to perform better than the GCP-enhanced model.
• The reduced model, which relies on significantly less data than the full model to estimate the underlying parameters, begins to show subtle signs that suggest the relationship between GCP data, sentiment, and market returns may be more complex and, consequently, less stable than in the control model. This complexity was discussed in Holmberg (2024) as a potential driver of the results, and these initial findings appear to support that hypothesis.It will be interesting to continue tracking these trends as more data becomes available.

2025-08-08

There was a decline in global risk sentiment yesterday due to renewed trade-related worries that emerged after the Swedish market had closed. This led to an intraday spike in the VIX (which later moderated), a rise in risk premiums, and a moderate drop in S&P 500 valuations (-0.08%). Since the daily forecasts could not account for these developments, both Model A (control) and Model B (GCP data-enhanced) issued a buy signal. Nonetheless, thanks to its past performance, the GCP-data-enhanced model (Model B) continued to outperform the control model (Model A).

Full Model

• Total Return: Model B has currently delivered a return of 3.25%, compared to 2.66% for Model A.
• Hit Rate: Model B correctly made the correct trade 60.00% of the time, versus 56.67% for Model A.

Reduced Model
Though performance is lower than under the full specification, the GCP‑enhanced model still shows clear benefits:

• Total Return: Model B returned 0.90%, versus 0.53% for Model A.
• Hit Rate: Model B achieved 53.33%, compared to 50.00% for Model A.

Benchmark Comparison
The buy-and-hold (B&H) benchmark, which is not subject to trading costs, has yielded a return of 3.24%, slightly lower than the GCP data-enhanced full model. Its directional accuracy (i.e., proportion of time long) is 41.94% since the simulations began, underperforming both GCP-enhanced configurations in terms of directional accuracy.

2025-08-07

The GCP-data-enhanced model (Model B) continues to outperform the control model (Model A) under both the full and reduced specifications.

Full Model

• Total Return: Model B has delivered a return of 3.81%, compared to 3.22% for Model A.
• Hit Rate: Model B correctly predicted market direction 62.07% of the time, Model A 58.62%.

Reduced Model

Although performance is lower than under the full specification, the GCP-enhanced model still shows improvement:

• Total Return: Model B returned 1.45%, versus 1.08% for Model A.
• Hit Rate: Model B correctly predicted market direction 55.17% of the time, Model A 51.72%.

Benchmark Comparison

For reference, the buy-and-hold (B&H) benchmark during the same period yielded a return of 3.32%, with 43.33% of trades resulting in positive returns.

2025-08-06

Currently, the simulations show that the GCP-data-enhanced model (Model B) consistently outperforms the control model (Model A) under both the full and reduced specifications.

Full Specification

Under the full specification, Model B delivered a total return of 3.23%, outperforming Model A’s 2.64%, resulting in a gain of +0.59 percentage points attributable to the GCP data. In terms of predictive accuracy, Model B also showed a stronger hit rate of 60.71%, compared to 57.14% for the control model—an improvement of +3.57 percentage points.

Reduced Specification

In the reduced specification, Model B again outperformed, generating a total return of 2.03%, versus 1.65% for Model A, amounting to a +0.38 percentage point gain. Similarly, the hit rate improved from 53.57% to 57.14% with the inclusion of GCP data, once again reflecting a +3.57 percentage point gain in directional accuracy.

For comparison, the buy-and-hold (B&H) benchmark over the same period yielded a return of 2.58%, with 41.38% of trades resulting in positive returns.

-------------------------------------------------------------------------------------------------------------------------------------

This section outlines the ongoing one-day-ahead econometric simulations comparing two preregistered models—one incorporating Global Consciousness Project (GCP1) data and one excluding it. The framework extends the analysis presented in the 2024 study, testing whether the GCP-linked signal continues to provide predictive value in real-time market conditions through 2025.

In the 2024 study, statistically significant correlations were observed between GCP1 data (specifically the Max[Z] metric) and financial market movements. That study also included a one-year out-of-sample simulation to validate the findings, with supportive results. This new study builds on that work by extending the analysis into a new one-day-ahead simulation framework running through the end of 2025. The goal is to evaluate the stability and practical predictive potential of GCP1 data, and to further assess the proposed GCP data linked metric beyond the original simulation horizon.

New econometric models have been developed using EViews 14, and all models have been preregistered to ensure transparency and replicability. Full model specifications are available at: https://osf.io/jk24c.

The data used for the study can be found here: OSF | Data

The EViews code to replicate the preregistered tables can be found here: OSF | Data

Updatas on the simulations can be found here.

Initially, the active strategies derived from the econometric models calculated returns using the opening price as the entry cost of the contract. However, since the timing depends heavily on investors’ risk aversion, the opening price has been ignored since 4 September 2025. Once the results are finalized, prices at different times during the morning will be used to evaluate the strategies’ performance.

It is noted that as the number of active RNG devices contributing to the GCP1 network continues to decline, the strength of the signal may weaken which could potentially make the results more fragile over time. This is acknowledged and will be monitored throughout the simulation. Still, the outcomes presented here, together with those from the original publication, are intended to guide the development of a more refined modeling framework once GCP2 data becomes readily available.

The infrequent updates presented here compare the performance of the preregistered models, which differ only in whether they incorporate the GCP1-based Max[Z] signal:

• Model A (Control): excludes GCP data entirely
• Model B (GCP): incorporates the Max[Z] metric from the Global Consciousness Project (GCP1)

The performance of these models is also assessed under two sample horizons:

• Full Model: uses all available data from 1999 to 2025
• Reduced Model: uses only data from the beginning of 2025 onward

Model performance is evaluated across two key metrics:

• Total Return: the overall return generated by the strategy implied by the model
• Hit Rate: the percentage of correct directional predictions (buy vs. sell decisions)