Stress testing a strategy against flash crashes: a step by step method
A flash crash is a brief, sharp drop in prices that lasts minutes, sometimes seconds, and often reverses just as quickly. For retail traders in Sydney or Melbourne who watch markets after the morning commute, the drama usually unfolds in the overnight window when liquidity is thin. Defining these events precisely matters, because a method built to test for routine losing days will not necessarily catch a five-minute waterfall that wipes out a month's gains. Once the definition is clear, the stress-testing workflow becomes a repeatable weekly exercise rather than a panic response to a single event.
Stress testing answers a single question: how would my current rules behave if something similar happened again? It does not predict the future, but it does expose the assumptions hidden inside backtest equity curves, especially assumptions about fills, spread, and the willingness of a trader in Perth or Brisbane to actually click confirm when the screen turns red. The method below is platform-agnostic and works for equities, FX majors like AUD/JPY, and the more volatile parts of the crypto market.
Defining flash crashes in a trader's terms
A flash crash is not the same as a bear market or a routine gap-down on the ASX at open. It is a liquidity event: bid-ask quotes evaporate, stop orders cascade, and prices move further than fundamentals justify. The May 2010 US equity flash crash, the January 2015 Swiss franc removal of the EUR/CHF floor, and the May 2021 crypto liquidation cascade all share this signature. For Australian traders, the relevant ones often arrive during the Asia session, when the local day is just beginning and order books are still building depth.
What separates a flash crash from a slow bleed is the speed of recovery. If the move persists for days or weeks, the strategy should already be designed to handle that through position sizing and stop rules. A flash-crash scenario is specifically about hours, minutes, or seconds of dislocation followed by partial or full reversion. Building a stress test around anything longer misses the point.
Building a scenario library that fits your markets
A useful library mixes historical and synthetic events. Historical events ground the test in real tick data, which is essential because synthetic candles will not reproduce the slippage patterns of a real liquidity vacuum. Pull tick or one-minute bars from at least three reference events: a US session cascade (such as the August 24, 2015 ETF-related moves), an Asian session event (the August 2015 RMB devaluation that hit AUD crosses hard), and a crypto-native event (March 12, 2020, often called Black Thursday). These three cover equities, FX, and digital assets.
Synthetic scenarios fill the gaps. If you trade AUD/JPY from Australia, design a Tokyo-open scenario where the pair gaps 200 pips against you while your local broker is still in the pre-open. If you hold altcoins, design a scenario where a major listing triggers a -30% retrace within ten minutes, a common pattern that reactive traders tend to chase. Save each scenario as a reusable file with documented assumptions so the test is repeatable and auditable later.
Wiring scenarios into your backtest and replay tools
Most backtesting platforms allow you to import custom price series or replay bar-by-bar. The point is to feed your strategy the same decision tree it would see live: entry signals, position sizing, stop placement, and exit rules must all fire on the replay data exactly as they would on a real Tuesday afternoon. Anything less is theatre. Where the platform cannot replay tick-by-tick, a spreadsheet-based simulator using the closing prices plus modelled slippage is acceptable as a second-best option.
Slippage is the most underestimated input. During flash crashes, spreads widen five to fifty times their normal value, and many retail orders are rejected or partially filled. Build at least three slippage assumptions: a conservative base (1.5× normal spread), a stressed level (5×), and an extreme level (15×) for worst-case modelling. The cost difference between these levels is where most strategies quietly fail. Pair this with clear daily and weekly entry caps so the system itself refuses to add risk around scheduled volatility windows, as outlined in this article on overtrading rules.
Document every assumption in a single sheet before each run. If the assumption is not written down, it will quietly drift between sessions and your comparisons will become meaningless. The discipline of pre-recording is what separates a stress test from a hopeful backtest.
Reading the stress-test report
Once the replay finishes, the report should show three numbers per scenario: peak drawdown in account percentage, time to recover the original equity high, and the worst single fill price your strategy actually received. If your equity curve looks smooth except for one V-shaped dent, that is a sign your stop rules worked but your slippage assumptions were too generous. If the curve has a stair-step down across every scenario, the problem is structural and no amount of stop tightening will save it.
Cross-reference the report against the assumptions. If a scenario causes a 12% drawdown under conservative slippage but a 28% drawdown under extreme slippage, the strategy is highly sensitive to execution quality, which means it should be traded only through venues with reliable fills. Avoid the temptation to relabel losing days as liquidity events after the fact; that habit, common in reactive crypto trading, is the same pattern that drives traders to chase listings they did not research, a behaviour this altcoin listing guide unpacks in detail.
Comparing the scenarios side by side is the easiest way to see which conditions genuinely threaten your approach. Below is a summary of five widely studied flash events and the lesson each one tends to teach a disciplined retail trader.
| Event | Date | Market | Approx. drop | Time to partial recovery | Key lesson for retail traders |
|---|---|---|---|---|---|
| US equity flash crash | 6 May 2010 | US equities | ~9% intraday | ~30 minutes | Circuit breakers matter; stop cascades amplify |
| EUR/CHF floor removed | 15 Jan 2015 | CHF pairs | ~30% intraday | Years for some pairs | Central bank policy is not a stop loss |
| RMB devaluation | 11 Aug 2015 | Asia FX | 3-4% on AUD crosses | Hours | Asia-open gaps hit AUD positions hard |
| Crypto Black Thursday | 12 Mar 2020 | Crypto majors | ~50% BTC | ~6 months | Liquidity, not valuation, drives the floor |
| LUNA/UST collapse | 9-12 May 2022 | Altcoins | ~99% LUNA | Never | Stablecoin pegs can break; size accordingly |
Recommendations for stress-test hygiene
A stress test is only as useful as the discipline around it. Run the suite at fixed times, store the reports in a dated folder, and never edit the strategy between runs without rerunning every scenario. Versioning matters because the goal is to compare like with like, and a quietly shifted parameter will corrupt every comparison you try to make later. Treat the library as living data, not a one-off project, and add a new scenario after every real-world event that surprised you.
A practical set of habits worth keeping:
- Run the full scenario library every month, not just after a real crash, since the value is in routine rather than reaction.
- Re-run after any code change to entry or exit logic, even a one-line edit, before going live with the new version.
- Keep slippage assumptions in a separate file from the strategy so they can be revised without touching core logic.
- Record your emotional state in a one-line note after each run; patterns in your own reactions are part of the data.
- Cap any single strategy at 2% account risk per trade, because no stress test covers every tail event.
- Review the worst single fill from each scenario, not just the headline drawdown, since fill quality often reveals more than the equity curve does.
Stress testing is less about predicting the next waterfall and more about knowing in advance what your screen will look like if one arrives. The trader who has already seen their strategy lose 8% on replay data is far calmer than the one seeing it for the first time with real money on the line. The worst single fill from a single weekend replay is the kind of number that belongs in the trading journal before Monday's open, because it converts abstract risk into a concrete figure the rest of the system can be sized around.