Education
Contents
By Arizmic LLC5 min readPublished Last reviewed

Performance paths and trade distributions

Read the route a strategy took, not only the summary statistic at the destination.

The path explains the headline

A total return tells where equity ended. It does not show how the strategy got there, how much capital was at risk, or whether the result depended on a few unusual trades.

The performance path is the sequence of equity, exposure, gains, losses, and drawdowns through time. The trade distribution is the collection of individual or grouped outcomes behind that path. Reading both can turn “the strategy made 20%” into an explanation of what actually happened.

This matters because many different processes share the same endpoint. A steady series of small gains, one enormous gain after months of losses, and a highly leveraged round trip can all end with the same total return.

Start with equity and drawdown together

The equity curve shows cumulative wealth under the configured accounting. The drawdown curve shows the percentage decline from the prior equity peak.

Read them together:

  • A smooth equity curve with one sudden drawdown may indicate tail exposure.
  • Long flat periods may show regime dependence or capital tied up without return.
  • Fast recovery after repeated losses may still require substantial margin.
  • A late surge can dominate a full-sample metric and leave little independent evidence.

Also inspect exposure. A flat equity curve while the strategy is flat is different from a flat curve while it carries a large offsetting position.

Trade averages hide shape

Win rate is the proportion of winning trades. Average win and average loss describe magnitude. Expectancy combines frequency and size:

E=pwW‾−(1−pw)L‾E = p_w\overline{W} - (1-p_w)\overline{L}

Here, pwp_w is the observed win proportion, W‾\overline{W} is average winning amount, and L‾\overline{L} is the positive magnitude of the average loss.

Illustrative example. Suppose 80% of trades win 1 unit on average while 20% lose 5 units on average. Expectancy is 0.8(1)−0.2(5)=−0.20.8(1)-0.2(5)=-0.2 unit per trade. The high win rate and negative expectancy are both correct because loss magnitude overwhelms win frequency.

The reverse is also possible: a low win rate can accompany positive expectancy when gains are sufficiently larger than losses. Neither average describes tail events, clustering, or changing position size.

Group the trades by the decision that produced them

Useful breakdowns include:

  • long versus short;
  • instrument and market;
  • entry signal or setup;
  • exit rule;
  • holding period;
  • volatility or session state;
  • calendar period;
  • gross versus net cost;
  • ordinary versus largest outcomes.

If “62% of losing trades hit the same stop” or “one instrument produced 70% of profit,” the result suggests a specific research question. Without attribution, the researcher may change the wrong component.

Compare trades on a common economic scale

Currency profit and loss can make the largest positions dominate a trade distribution even when their decisions were ordinary. A 5,000 gain on a large position and a 500 gain on a small position cannot be compared meaningfully without their capital, risk, or exposure.

Useful scales include:

  • return on allocated capital;
  • basis points of notional;
  • profit or loss in units of the planned trade risk;
  • price movement per contract or share; and
  • contribution to portfolio return.

No scale is universally correct. The purpose is to separate decision quality from position size while retaining the original currency result for economic reconciliation.

Duration and overlap change interpretation

Trade counts are not automatically independent observations. A portfolio may hold several trades driven by the same market move. Daily strategies can enter multiple correlated instruments on one signal. Scaling in may create many rows from one economic position.

Report positions, episodes, and time periods where useful—not only raw trade count. A thousand intraday fills during ten market events do not provide the same evidence as a thousand independent decisions.

Costs should be visible in the distribution

Average net trade can be small even when gross trade expectancy looks healthy. Costs often vary with turnover, volatility, order type, and trade direction.

Inspect:

  • gross and net profit and loss (P&L);
  • fees and modeled slippage;
  • implementation shortfall where available;
  • cost as a fraction of gross edge;
  • whether losing trades pay systematically higher cost;
  • whether the largest positions occur when liquidity is worst.

A strategy that survives only with average costs applied uniformly may fail when costs are conditioned on the states in which it trades most.

A worked reading sequence

Illustrative example. Suppose a constructed run reports positive total return and Sharpe. Read it in this order:

  1. Confirm the dataset, Strategy version, window, execution fidelity, and costs.
  2. Check whether the equity curve rises throughout the sample or in one period.
  3. Locate maximum drawdown and recovery duration.
  4. Inspect trade distribution, outliers, and concentration.
  5. Break outcomes down by side, instrument, entry, and exit.
  6. Reconcile gross and net P&L.
  7. Identify which claim remains supported and which needs another test.

The process may reveal that the headline is valid but narrow—for example, “profitable in this sample because short trades during two trends offset a larger number of small long losses.”

Common mistakes when reading results

  • Jumping directly to net profit or Sharpe.
  • Treating trade rows as independent observations.
  • Ignoring flat time and exposure.
  • Reporting maximum drawdown without duration or recovery.
  • Removing an outlier without a rule that would have excluded it in real time.
  • Looking at gross expectancy while presenting net portfolio return.
  • Aggregating trades until exit attribution and market concentration disappear.

Further reading