Why Most Retail Trading Systems Fail

The regulator-disclosed numbers do not leave much room for interpretation. 74% to 89% of retail CFD accounts lose money, the mandatory ESMA disclosures have read since 2018. The FCA’s own analysis comes out at around 80% for UK accounts. CFTC enforcement records contain a long parade of “systems” that turned out to be nothing of the sort. These figures do not move year to year, which is the most informative thing about them: whatever the average retail trader is doing, it is structurally not working.

The structural reasons are well-understood. They are also unusually absent from the trading-system marketing material that beginners encounter first. This article walks the five that matter most, the arithmetic and evidence behind each, and what an honest reader can take from them.

1. Survivorship bias on the marketing side

The trading-education industry is staffed almost entirely by survivors. The systems that get sold are the systems whose authors are still in the market to sell them. The systems whose authors blew up, gave up, or moved on are not represented at all.

This sounds obvious. The implications are not. If you sample a hundred random retail traders in 2016 and run them forward to 2026, the regulator-disclosed loss rates suggest that 74-89 of them are out by the end window. Of the 11-26 still trading, a single-digit percentage will be doing well enough to make trading their primary income. Those few are the ones writing the courses, the YouTube channels, the trading rooms. They are the visible tip of a distribution whose body is completely hidden from the customer’s view.

The asymmetry is not malicious. It is structural: failure is silent, success is loud. But the asymmetry means that any “successful traders use this method” claim is almost definitionally selecting on the success outcome. You cannot infer the method’s average performance from the success cases, because the failure cases of the same method are silent in the same way.

2. Selection bias inside the backtests

The system-marketing flywheel runs on backtests. “This system would have turned $10,000 into $850,000 over the last decade.” The chart is usually green and smooth. The drawdowns are footnoted, if at all.

A non-exhaustive list of biases that produce inflated backtested numbers:

• In-sample optimisation. The parameters of the system (moving average lengths, RSI thresholds, stop distances) are usually tuned to maximise historical performance. The system would have produced $850,000 because the parameters were chosen with full knowledge of which parameters worked. Out-of-sample, the same system typically reverts to coin-flip behaviour.
• Survivorship in the data. Tests run on currently-existing pairs miss the pairs that were delisted, currencies that were redenominated, or markets that simply died. In FX this matters less than in equities, but in any backtest that touches commodities or exotic crosses, the bias is non-trivial.
• Look-ahead. Indicators that reference values at the close of a period are sometimes computed using data from after the entry decision. Backtests that get this wrong show improbably-good entries; live trading shows the realistic version.
• Transaction-cost neglect. Spreads, swaps, commissions, and slippage are routinely either zero or constant in a retail backtest. Live trading produces variable, occasionally large friction, particularly during the news events that the system also wants to trade. (See Trading Costs Explained.)

When an honest, out-of-sample, transaction-cost-aware backtest is applied to the same strategy, the equity curve usually flattens dramatically and often goes negative. The system that “would have made $850,000” turns out to break even, or worse, when the biases are stripped out.

3. The arithmetic of frequency

A second structural problem is that most retail systems run at frequencies whose friction cost exceeds any realistic edge. A day-trading system that takes four round-trip trades per day on EUR/USD at a typical 1.5-pip retail spread is paying $6 per day per mini lot. Across 250 trading days, that is $1,500 of friction per mini lot. On a $5,000 starting account, the friction alone is 30% of starting capital per year, before the system has its first losing trade.

A more honest way to compute viability before deploying any system:

Edge per trade (in dollars, after costs and slippage) × expected trades per year > target return + tail-risk buffer

For most retail systems, the per-trade edge is at best a few cents on a mini lot, the costs are an order of magnitude larger, and the inequality runs the wrong way. (See Risk Management Basics for the expectancy framework.)

4. Regime change

Markets do not stay in one regime. Strategies that worked in the 2010s low-volatility quantitative-easing regime did not work in the 2022-2023 inflation regime. Strategies that worked in the high-vol post-2008 period stopped working as soon as volatility compressed. Strategies that depended on the yen as a funding currency stopped working in August 2024 when the BOJ exited the zero-rate regime and the carry trade unwound globally.

A backtest, however clean, can only sample regimes that have already happened. The system fitted to past regimes is implicitly betting that the next regime will look like the average of the last ones, which is the regime nobody actually lives in. The strategy literature (Pesaran et al., among others) consistently finds that out-of-sample performance is dramatically worse than in-sample for any system tuned to a specific regime window.

For a retail trader, this means: the longer you backtest, the more honest the test, but the shorter the relevance window of the finding. A system that “worked for ten years” worked across a specific set of regimes that may not repeat. There is no setting of “more data” that fixes this.

5. The behavioural layer

Even systems with a small genuine edge can be ruined by the way real traders interact with them. The classic Barber and Odean line of research (2000, “Trading Is Hazardous to Your Wealth”; Barber et al. 2014 on Taiwanese day traders) finds that retail traders consistently underperform the markets they trade by margins larger than transaction costs alone would explain. The gap is behavioural:

• Overconfidence. Traders systematically overestimate their edge and undersize positions when they should be cautious.
• The disposition effect. Winning trades are closed too early (“locking in profit”); losing trades are held too long (“it has to come back”). The effect compresses winners and stretches losers, which is structurally bad for any system with positive expectancy.
• Revenge trading. A loss triggers a larger, lower-quality next trade. The arithmetic compounds in the wrong direction.
• Recency bias. Performance over the last few trades dominates the trader’s sense of the system, so a long-expectancy system is abandoned during the streaks the binomial distribution guarantees.

The prospect-theory literature (Kahneman and Tversky 1979 and the large body that followed, including more recent work on prospect-theoretic preferences in online trading like arXiv:1402.6393) provides the formal framework. A system with even a modest positive expectancy can be turned into a losing strategy by the trader’s discretionary deviations from it. The system did not fail. The execution did. From the outside, the two are indistinguishable.

What survives

This is not a counsel of despair. A small minority of retail accounts do persist and produce positive returns over multi-year windows. What distinguishes the survivors, in the academic literature and in the more honest practitioner accounts:

• Modest, well-understood edges. Survivors rarely claim spectacular returns. The honest expectations are 5-15% per year after costs, with meaningful drawdowns along the way.
• Strict position sizing. A fixed-percent risk per trade, applied without exception, sized for survival rather than optimal growth. See Risk Management Basics.
• Cost-aware design. Frequency is set by the friction arithmetic, not by how often the trader wants to be in the market.
• A real thesis. The trade is taken because something has been observed about how the market works (a rate-differential story, a positioning extreme, a structural flow), not because a chart pattern triggered.
• Acceptance of long flat periods. The survivors do not need to be active every week. They sit out regimes that do not suit their approach.
• No system change after losses. The behavioural literature is clear that abandoning a positive-expectancy system during a losing streak is the single most common path to underperformance. The survivors stick to the plan, sized small enough that the streaks the math guarantees are survivable.

This list looks unglamorous. That is the point. The trading-system marketing pipeline sells excitement, and excitement is exactly what gets traders into the 74-89% loss-rate bracket. The structure of what works is dull on purpose.

The takeaway

Most retail trading systems fail for structural reasons that do not depend on the specifics of the system: survivorship bias on the marketing side means failure cases are invisible; selection bias in backtests inflates expected returns; transaction-cost arithmetic at retail frequencies eats most apparent edges; regime change makes any one-window fit a poor predictor of the next window; and the behavioural layer turns even genuine edges into losses when the trader cannot follow the plan.

The structural understanding of why systems fail is the prerequisite for evaluating any system you encounter. A claim that sidesteps these mechanisms is either incomplete or selling something. A claim that engages with them honestly, and shows the arithmetic, is a useful starting point. The first claim is more common than the second by a wide margin.

For the academic record on what does work in FX, the next strategy pieces cover trend following and carry trades and the unusual asymmetric properties each carries.