Indicator Deep-Dive: Where Our Signals Work, Where They Don't, and Why the Gap Is Larger Than Expected

The Question Traders Should Be Asking Before They Apply Any Indicator

Most retail traders treat technical indicators as universal tools. RSI divergence on a forex pair, a moving-average crossover on a commodity futures contract, a momentum signal on an individual stock — the assumption, often implicit, is that if the indicator works somewhere, it works everywhere. Our data from the past signal cycle suggests that assumption is not just imprecise. In several asset classes, it appears to be actively harmful. This report examines 32 closed signals across four asset categories — crypto, forex, stocks, and commodities — and asks a narrower, more answerable question: does indicator performance vary systematically by asset class? The answer, with all appropriate caveats about sample size, is yes. The variance is large enough to matter for position sizing, category selection, and, frankly, whether you should be running these signals at all in certain markets right now.

The Headline Numbers Hide a Four-Way Divergence

The aggregate figures are, on their face, discouraging. Across all 32 closed signals, the overall win rate stands at 37.5% with an average trade return of -0.51%. A system losing money on average, winning less than four trades in ten — that is the headline. But the aggregate is doing what aggregates often do: it is averaging together genuinely different phenomena and producing a number that accurately describes none of them.

The table below breaks down performance by category.

Read across that table slowly. Crypto delivered a 100% win rate and an average return of +2.96% per signal. Commodities delivered a 16.7% win rate and an average loss of -2.19% per signal. That is not noise around a shared mean. That is a structurally different outcome in each category, and it demands a category-by-category explanation rather than a single system-wide verdict.

Forex: The Only Category with a Defensible Sample

Forex is the only asset class where we have enough observations to say something with moderate confidence. With 16 signals, it remains well below the threshold where we would use the phrase "statistically significant" — but it is the closest we have to a representative sample in this dataset. The results are, in a word, neutral. A 43.75% win rate combined with an average return of -0.04% per trade describes a system that is essentially breaking even on transaction costs and generating no alpha. That is not a crisis, but it is not a thesis either.

What the forex numbers suggest is that the indicators being applied — and without breaking out by specific signal type, we cannot go further than this — are neither well-suited nor poorly-suited to currency markets in the current regime. The win rate of 43.75% is below the 50% breakeven threshold for a symmetric payoff system, but the near-zero average return implies that when wins occur, their magnitude is roughly proportional to losses. This is the signature of a system that is not misfiring; it is simply not finding an edge. Whether that is a function of the indicators themselves, the macro regime in currency markets during this period, or some interaction of both is a question this data cannot answer.

Forex also carries the methodological benefit of being the most liquid category in the set. Slippage is lower, fills are more reliable, and the realized returns are more likely to reflect the signal's actual predictive value rather than execution friction. That makes the neutral result here more credible than the same neutral result would be in a thinly traded instrument.

Stocks and Commodities: Where the System Is Losing Money

The stocks and commodities numbers are where this report gets uncomfortable, and discomfort is worth sitting with. Eight stock signals produced a 25.0% win rate and an average return of -1.05%. Six commodity signals produced a 16.7% win rate — meaning roughly one in six signals was a winner — and an average loss of -2.19%. These are not rounding errors. A system that wins one in six trades and loses two percent on average per trade is a system that needs to either be suspended in that category or substantially recalibrated.

The commodities result is particularly worth unpacking, even if the sample is too small to be definitive. A 16.7% win rate is consistent with a system that is directionally wrong in the majority of its calls. In commodity markets, this can happen for several structural reasons: mean-reversion signals may be fighting carry dynamics or supply-side event risk; momentum signals may be lagging against faster-moving physical market participants; or the volatility regime in commodities may simply be incompatible with the lookback periods and thresholds used to generate signals. We do not have enough granularity in this dataset to distinguish between those explanations. What we can say is that the indicator framework, applied to commodities as a category, has produced negative expected value across the six closed signals we have observed.

Stocks at -1.05% average and 25% win rate fall between the forex and commodities extremes, but the direction is unambiguous. Individual equity signals are performing worse than forex signals on both dimensions. One plausible explanation is that single-stock signals carry idiosyncratic risk that aggregate indicator models are not designed to handle. A relative-strength signal on an index might behave predictably; the same signal applied to an individual stock can be overwhelmed by an earnings release, a sector rotation, or a single analyst call. Whether the solution is to shift stock signals toward index instruments, tighten stop parameters, or exit the category entirely is a portfolio-management question — but the data provides the starting point.

The Crypto Number Is Real, But You Cannot Build a Strategy on It

Two signals. One hundred percent win rate. Average return of +2.96%. The crypto numbers are the most striking in the dataset and the least actionable. With n=2, this is anecdote, not evidence. Two winning trades in crypto is entirely consistent with random outcomes; it tells us nothing reliable about whether the indicators being applied have genuine predictive power in digital asset markets. We are stating this explicitly because the temptation to over-index on the best-performing category is a well-documented behavioral bias, and our job is to counteract it rather than feed it.

What we can say, narrowly, is that the two signals generated during this period captured meaningful directional moves — a combined return implying moves of roughly that magnitude in the underlying instruments. Whether that reflects skill in indicator design, a favorable volatility environment in crypto during the measurement window, or simple luck is unknowable at this sample size. We need substantially more observations before we assign the crypto result any weight in a forward-looking strategy allocation.

Where the Data Does Not Support a Confident Conclusion

Intellectual honesty requires a section on what this report cannot claim. There are several things the data above does not tell us, and treating these gaps as answered would be a methodological failure.

• We cannot identify which indicators are driving the divergence. The data is broken out by asset class, not by indicator type. Whether the commodities underperformance is driven by a specific signal — say, a particular oscillator or a breakout setup — or whether it is spread evenly across the indicator suite is unknown. That distinction matters enormously for what corrective action to take.
• We cannot isolate regime effects from structural effects. If equity markets experienced a specific directional bias during the measurement window that happened to run contrary to our signal positioning, the stock results could be a period effect rather than a structural failure. The same logic applies to commodities. We need a longer time series across multiple market regimes to separate these explanations.
• No category in this dataset clears the threshold for statistical significance. Forex at n=16 is the closest. Stocks at n=8, commodities at n=6, and crypto at n=2 are all preliminary observations. We are describing patterns in small samples, and patterns in small samples often disappear when the sample grows. This is preliminary. We are publishing it because directional transparency is more useful than waiting for a dataset large enough to be conclusive — but readers should weight these findings accordingly.
• We have not controlled for position sizing or holding period. An average return of -2.19% in commodities could reflect a short holding period with tight stops or a longer hold that drifted further negative. The return figures alone do not capture the risk-adjusted story.

What Traders Should Actually Do With This

Despite the caveats, the data has enough shape to support several practical decisions. These are not recommendations to trade specific instruments; they are framework adjustments for how to interact with our signal output by category.

• Reduce or suspend commodity signals pending further data. A 16.7% win rate and -2.19% average return across six closed signals is a clear negative signal-about-the-signal. Traders following commodity alerts should either reduce position size materially — think 25-50% of normal sizing — or wait for a larger sample that shows improvement before resuming standard allocation. The expected-value calculation does not support full sizing at current observed rates.
• Treat forex signals as breakeven infrastructure, not alpha generation. A 43.75% win rate and -0.04% average return suggests forex signals are not currently earning their risk. They may be useful for hedging directional exposure or maintaining engagement with a liquid market, but traders expecting consistent positive returns from this category in the near term should revisit that expectation. If transaction costs are even modestly elevated — as they can be in less liquid currency pairs — the neutral average return turns negative in practice.
• Audit stock signals for idiosyncratic risk exposure. The 25% win rate in stocks suggests the signal framework may be poorly calibrated to single-name dynamics. Traders using stock alerts should cross-reference signal timing against scheduled earnings, index rebalancing dates, and sector-rotation flows. A technically valid signal that runs into a binary event is not a failure of the indicator — but it is a failure of the filtering layer around it.
• Do not increase crypto allocation based on two data points. The temptation is understandable. The discipline required to resist it is part of systematic trading. Wait for n≥20 in crypto before treating that category's results as a meaningful input to sizing decisions.
• The overall system return of -0.51% should prompt a review, not a shutdown. A single measurement window at -0.51% average across 32 signals is not a catastrophic failure. It is a yellow flag. The appropriate response is disaggregation — which is what this report attempts — followed by targeted adjustments in the underperforming categories, not wholesale abandonment of the framework.

Methodology Note

This analysis is based on 32 closed signals generated and tracked through the Stocks365 proprietary backtest and live signal system as of April 20, 2026. The dataset covers four asset categories: crypto (n=2), forex (n=16), stocks (n=8), and commodities (n=6). No category in this sample meets conventional thresholds for statistical inference; the largest subcategory, forex, contains 16 observations. All return figures represent average percentage return per closed signal within each category. We have not applied any risk-adjustment, Sharpe-equivalent normalization, or drawdown weighting to these figures — the numbers presented are simple averages of closed-trade returns as reported by the system. Holding periods, entry and exit methodology, and specific instruments within each category are not broken out in this dataset. The findings in this report should be treated as directional and preliminary. They are published in the interest of transparency and as a foundation for ongoing monitoring, not as a definitive performance assessment. We will revisit these conclusions as the signal count grows.