Why Do Traders Lose Money With a Profitable Strategy?

The setup wins often enough. The backtest looks sensible. The account keeps drifting lower.

Before replacing the strategy, compare the strategy you tested with the trades you actually placed. They may have different costs, position sizes, entry prices, exits and even market conditions. A small edge leaves very little room for those differences.

A positive strategy expectation can become a negative account result through execution costs, inconsistent sizing, deviations from the rules or ordinary sampling variation. There is also a less comfortable possibility: the apparent edge was never reliable outside the test sample.

Put the edge into units you can audit

Let one R mean the dollar loss planned for a trade if its intended stop fills as assumed. A $100 planned loss is 1R; a $180 gain is 1.8R. This notation makes trades easier to compare, provided you define R at entry and do not quietly change it afterward.

Suppose an illustrative strategy wins 45% of trades. Winners average 1.8R and losers average 1R. Its gross expectancy is:

0.45 × 1.8R − 0.55 × 1R = +0.26R per trade

That is an average under the assumed distribution. It is not what the next trade owes you. Subtract 0.10R in average costs and the expectation becomes +0.16R. Another 0.20R of average execution shortfall turns it negative.

Costs are larger when the planned move is small

Trading fees, spread, slippage and funding consume part of the expected payoff. Count entry and exit costs. If your analysis already includes slippage, do not subtract it again under an execution adjustment.

In a fictional $10,000 position with a $100 planned price risk, $10 of total costs consumes 0.10R. Keep the same notional and costs but tighten the price risk to $50, and costs become 0.20R. The fee schedule did not change. The strategy's tolerance for costs did.

Barber and Odean's study of individual equity investors found that frequent trading was associated with poorer net performance in their brokerage-account sample. It concerns historical stock investors, not present-day crypto terminals. Its useful lesson here is to examine net results rather than assume extra activity creates value.

Position sizing changes the strategy you are running

A list of wins and losses is not enough to reconstruct account performance. You also need the size assigned to each trade.

Imagine three outcomes of −1R, −1R and +2R, using a fixed reference R of $100. At equal size, the sequence breaks even before costs. If the first two trades each risk $200 and the last risks $50, the result is −$300. The same directional calls now produce a different account history.

Sizing often changes for reasons that never entered the backtest: confidence after a win, urgency after a loss, or discomfort after a drawdown. Measure those changes rather than calling the entire discrepancy bad luck. Keep planned dollar risk, actual quantity and account equity in the journal.

This is also where correlated positions matter. Five individually small bets on assets that move together can behave like one large bet. A per-trade limit does not describe the whole account's exposure.

The exit in the test must survive contact with the session

A backtest can take every stop without hesitation. A person can move one, cancel one or decide that this particular loss deserves more time.

Odean's research on the disposition effect examined brokerage records and found a preference for realizing gains over losses. That finding does not explain every losing account, but it gives you a concrete pattern to check: did live losers last longer or become larger than the strategy allowed?

Audit winners too. Closing profitable trades early can reduce the average winner enough to erase a payoff-dependent edge. A high win rate can coexist with poor expectancy if the few losses are large or costs are persistent.

Compare planned and actual entries and exits trade by trade. Include missed signals and skipped trades where records exist. Looking only at executed orders can hide selective participation in the strategy.

A profitable backtest may still be a weak claim

Start with the information that was available at each decision time. Revised datasets, future constituents, candle-close prices used before the candle closed and unrealistic fills can all make a simulation easier than live trading.

Then inspect the sample. A strategy selected from many variants has benefited from that selection process. Test the frozen rules on data that did not choose them. Keep the unsuccessful variants in your research notes so the winning chart is not presented as the only idea you tried.

Even honest rules with positive expectation can lose over a finite run. For a simplified independent process with a 55% chance of losing each trade, a particular block of four trades has a 0.55⁴, or about 9.15%, chance of containing four losses. That is not the probability of at least one such streak across a long trading history, and real outcomes need not be independent.

Build a short execution audit

For the next review, separate these records:

Beneat Terminal connects Hyperliquid and Binance accounts and provides order execution, position management and trading analytics. Its trading documentation describes the order and risk controls available in the interface. Use the records to compare decisions with the plan, rather than treating a dashboard score as proof of profitability.

If losses trigger unplanned entries, begin with the revenge-trading checklist. If session quality changes with fatigue, the sleep research guide explains what can be measured without inventing a biological explanation for every bad trade.

The first useful repair is the one tied to a documented gap. “Costs consumed half the tested edge” is actionable. “I need a better mindset” is much harder to test.