⏱ 1-Minute Summary A backtest is a statistical reference, not a prediction. Five gaps can never be fully closed: slippage and execution assumptions, dynamic margin and financing costs, liquidity and spread distortion, black-swan and dynamic events, and statistical and psychological bias. Knowing these limits is part of being a disciplined trader, and discipline is risk management: ignoring these gaps is how positions get out of control.
1. Slippage and Execution Price Assumptions
Backtests usually assume you can trade exactly at the closing price, or at the mid of bid/ask. In reality:
- Bid/ask spreads widen in stress; your fill is rarely at the ideal mid.
- Slippage on larger orders or illiquid strikes can be material.
- Timing: the closing price is a single point; your actual execution can differ.
Here is a concrete example. Suppose you want to sell 10 contracts of the $210 call, quoted $2.50 bid / $2.60 ask (mid $2.55). A backtest fills you at the mid, $2.55. In live trading you have to cross the spread, so as a seller you receive roughly the bid side and, as the quote moves, net around $2.45 per contract. That is about $0.10 worse per contract: on 10 contracts (1,000 shares) it is a $100 slip on a single fill, often larger than the position's theta for the whole day. Repeat that across 100 round trips and it is a silent $10,000 drag, enough to turn a marginal strategy from profitable to losing.
The fix is not to eliminate slippage (impossible to model perfectly) but to assume realistic costs and treat results with a buffer.
2. Dynamic Margin and Financing Costs
Backtests often hold margin and financing costs static, but in live trading they change:
- Margin requirements move with volatility, concentration, and broker policy.
- Financing / borrow costs vary over time and by symbol.
- A strategy's edge can shrink or vanish once realistic margin and funding are applied.
Consider a naked short straddle: at 20% IV it might need about $8,000 of margin; if IV spikes to 60%, your broker can raise the requirement to about $15,000. A position sized for $8,000 now needs nearly double the capital, forcing you to cut at the worst time. Borrow costs are just as easy to miss: for hard-to-borrow symbols, the borrow fee can reach 50–100% annualized, so a backtest that ignores it can make a short position look profitable when it actually loses money.
Long-term backtested returns are typically slightly higher than what is achievable live, in part because these costs are not accounted for.
3. Liquidity and Spread Distortion
A backtest uses whatever price is recorded; often far OTM strikes have thin liquidity and distorted IV. In live trading:
- You may not be able to fill the exact strike/expiry at the backtested price.
- Wide real-world spreads eat into edge that the data implied was there.
- Deep OTM or far-dated contracts can be nearly untradeable at size.
Here is a realistic case: on a stock at $200, a deep out-of-the-money $250 call shows a last recorded price of $0.10, but the actual market is $0.05 bid / $0.15 ask. The backtest assumes you sell at $0.10; live, the best you can get as a seller is the $0.05 bid. Half of the apparent "edge" was never actually tradeable.
The cleaner the data, the smaller this gap, but it never fully disappears.
4. Black Swans and Dynamic Events
Backtests cannot predict the unpredictable:
- Black swans: crashes, flash crashes, geopolitical shocks: are rare, violent, and absent from the fitted sample.
- Earnings and macro events (FOMC, product launches) cause gaps that EOD snapshots cannot reconstruct.
- A strategy that never hit a crisis in the backtest has simply not been tested against one.
The history is full of them: on 2010-05-06 the Dow fell about 1,000 points in minutes before recovering (the Flash Crash); 2008-09-15 was Lehman Brothers' bankruptcy; in March 2020 the pandemic produced one of the fastest ~30% drawdowns on record; and on 2020-04-20 WTI crude settled at −$37.63. None of these were in the fitted sample before they happened: a backtest that sailed through calm years was simply never tested against a shock.
This is a structural limit of historical data, not a flaw you can tune away.
5. Statistical and Psychological Bias
Finally, the human and statistical layer:
- Overfitting: optimizing parameters until the backtest curve looks perfect often fits noise, not signal.
- Selection bias: choosing the symbols or periods where the strategy "worked" inflates results.
- Psychological drift: real money changes behavior; discipline in a backtest is not the same as discipline at a live terminal. Bad discipline is bad risk management, the two failures reinforce each other.
The failure modes are common. Tweaking 12 parameters on a 3-year backtest until the equity curve looks perfect usually fits noise, and the "edge" mostly fails out of sample. Backtesting only 2017–2019 to "prove" a long-call strategy ignores the 2020 and 2022 drawdowns. And in live trading, most traders cut or add at the bottom of a 20% drawdown, while the backtest just clicks.
The result: even a statistically clean backtest overstates your likely live experience.
The Bottom Line
| Gap | What backtest misses | How to respond |
|---|---|---|
| Slippage & execution | Real fills ≠ closing price | Model costs; size with buffer |
| Margin & financing | Static assumptions | Stress with realistic costs |
| Liquidity & spreads | Thin-market distortion | Prefer liquid strikes/expiries |
| Black swans & events | Unpriced shocks | Expect the unexpected; diversify |
| Statistical & psychology | Overfit / selection / emotion | Out-of-sample tests; position limits |
A backtest tells you what happened under assumptions: it never tells you what will happen. Treat every result as a hypothesis to be validated, never a promise.
⚠️ Platform Data Boundary: This article discusses the limits of backtesting in general. This platform's backtests are based on T-1 EOD closing snapshots, which are well suited to multi-day-to-multi-month strategies but inherit all five limits above, especially around earnings, 0DTE, and intraday events that EOD data cannot reconstruct.
⚠️ Research Use Only: This article is educational. Backtesting cannot guarantee future results. Nothing here is a buy or sell signal. Use this platform's backtests as historical statistical reference only.