A clean equity curve is easy to fall in love with. The harder question is whether the strategy could have produced those results in the real world.
Before trusting a backtest that turns $10,000 into $2 million, check for three common problems: look-ahead bias, survivorship bias, and overfitting. Any one of them can make an ordinary strategy look exceptional.
1. Look-ahead bias
Look-ahead bias happens when a strategy uses information that would not have been available at the time of the trade.
For example, imagine your entry signal is calculated using Monday's closing price, but the backtest also assumes you bought at Monday's close. That is not realistic. You only know the final closing price once the bar has finished.
A more honest backtest would enter at Tuesday's open, or at some point afterward, once the signal could actually have been acted on.
This may sound like a minor timing issue, but it can completely change the results. Same-bar execution often gives a strategy unrealistically good entries, especially around reversals and short-term dips.
TensaFi's TestaFi engine uses next-bar-open fills by default to avoid this problem.
2. Survivorship bias
Suppose you backtest a strategy over the past 30 years using today's S&P 500 companies.
The result will probably look better than it should.
Today's index naturally contains companies that survived, grew, or remained important enough to stay included. It leaves out many of the businesses that failed, were delisted, went bankrupt, or were removed after years of poor performance.
In other words, you are testing the strategy on a list of survivors.
The proper solution is to use point-in-time index membership: the actual companies that belonged to the index on each historical date. Unfortunately, that data can be difficult or expensive to obtain.
A more accessible alternative is to test a single asset with a long, continuous history. Just be careful not to claim that a result from one ticker represents the entire market.
3. Overfitting
The more settings you allow yourself to adjust, the easier it becomes to create a strategy that looks perfect on historical data.
Twenty indicators with several thresholds each can produce millions of possible combinations. One of those combinations will probably look incredible, even if there is no real edge behind it.
That does not mean you discovered a brilliant strategy. It may simply mean you searched long enough to find the rules that best explain the past.
The strongest defense is out-of-sample testing.
Build and adjust the strategy using one portion of the data, then test it on a separate period that was not used during development. A common approach is to train on the first two-thirds of the dataset and reserve the final third for testing.
If performance collapses on the untouched data, the original strategy was probably overfit.
There is also a simpler defense: use fewer rules.
A strategy built around two indicators and a small number of parameters is less likely to be fitting random noise than one containing dozens of conditions. Simple does not automatically mean profitable, but it is usually easier to test, explain, and trust.
Quick reality checks
Some results should immediately make you skeptical.
A Sharpe ratio above 3 deserves a close inspection. It is not impossible, but it often points to overfitting, data leakage, unrealistic execution, or underestimated costs.
A strategy that goes years without a meaningful drawdown is also suspicious. Real strategies go through difficult periods.
Perfect entries near swing lows may indicate that future information is being used somewhere in the calculation.
And a strategy tested on only one ticker during one market environment has not been tested very thoroughly. Try it on several assets and across both bullish and bearish periods.
You should also include realistic assumptions for:
- Commissions and fees
- Bid-ask spreads
- Slippage
- Position sizing
- Liquidity
- Delisted securities
- Dividends and corporate actions
A strategy can survive clean historical prices and still fail once real trading frictions are included.
What a believable backtest looks like
A trustworthy backtest usually looks less exciting than a bad one.
It may have a Sharpe ratio around 0.8 to 1.5. It will probably experience noticeable drawdowns during difficult market regimes. Its rules should be simple enough to explain clearly, and the general behavior should remain similar across more than one asset or time period.
Most importantly, the results should make economic sense. You should be able to explain why the strategy might work, not just point to an upward-sloping chart.
TestaFi defaults to next-bar execution to reduce look-ahead bias. Pro also includes side-by-side sector comparisons, making it easier to spot strategies that only work on the single ticker they were designed around.
A backtest does not need to look perfect to be useful. In fact, the imperfect ones are often the most believable.