The COVID Stock Market Crash of 2020: Speed, Liquidity and Short-Trading Risk

A shock test needs an execution narrative

A rapid sell-off is not simply an ordinary downtrend compressed into fewer candles. The relevant research object is the sequence of quotes, signals, order submissions, accepted orders and fills. A chart can show a desirable entry while a real account receives a different price or no fill at all. Inspect the transactions rather than inferring execution from candle shapes.

For an invented example, a robot detects a SELL signal at 100 and plans a stop at 102. If its actual entry is 99 because the market moved before acceptance, the distance to the unchanged stop becomes three points instead of two. A position sized using the original distance now has a larger planned money risk. Recalculating after the fill, reducing the size or rejecting the trade are different policies that need their own implementation.

A gap can also carry the quote above the stop before a tradable price is available. Report that as execution risk. An account rule that requests closure at a threshold is not a guarantee that closure will occur exactly there. The stress experiment should record the requested action and the resulting action separately.

Test adverse conditions one at a time

Start with a reproducible baseline, then change one execution assumption at a time. Increase transaction costs, introduce supported delays or restrict eligible trading hours. Keep the other rules unchanged. This makes it possible to identify which condition explains a loss of performance rather than producing an opaque worst-case result with every variable changed simultaneously.

Next combine the most relevant stresses. A wider spread and a delayed fill can interact; their joint effect need not equal the sum of two separate tests. Label these variants as constructed stress cases. A test on historical ticks is historical replay, while deliberately changed costs and delays are scenario assumptions. Both can be useful, but they should not share a misleading label.

Check data continuity around exceptional sessions. Missing observations or synthesized prices can affect intrabar stops and trailing exits. Preserve the tester log and disclose what the dataset actually supports. A complete-looking report does not establish that every disruption was captured.

The recovery test is as important as the falling window

A short-only system can profit in the decline and then lose after the regime changes. Compare a report ending near the low with one that continues through recovery, without altering the preset. This exposes the dependence on a favorable ending date. Publish the continuation alongside the attractive section rather than hiding it in a separate file.

Measure the largest give-back after a profitable run. A robot may remain net profitable while returning most of its gains because its exit logic reacts slowly. Conversely, a rapid exit can stop losses but repeatedly close during ordinary countertrend moves. The appropriate choice follows measured trade-offs.

A meaningful pandemic case study would therefore include both the shock and its aftermath, with an exact contract, version and data range. The historical market story is an input to that experiment. It cannot stand in for a performance result from a robot whose trades were never recorded.