Plain-English definitions for every term EdgeAudit uses — strategies, statistics, and the bits in between. Written for someone who hasn't used a backtester before. Each entry has a definition, a concrete example, and links to related terms.
The total return of a strategy expressed as an equivalent per-year rate. If you made +21% over 2 years that's roughly +10% annualised. Lets you compare strategies tested over different windows on the same scale.
See also: Sharpe ratio, Buy and hold
Running a trading strategy against historical market prices to see what would have happened if you'd used it. Lets you test ideas before risking real money. The catch: most retail backtests are wrong in subtle ways — they look profitable in the data but lose money in real life. EdgeAudit's job is to apply the statistical discipline that makes a backtest honest.
Example. "Buy Apple when its 50-day average crosses above its 200-day average. Sell when it crosses back." A backtester walks through Apple's history, applies the rule, and tells you what would have happened.
See also: Strategy, In-sample / out-of-sample, Overfitting
A statistical adjustment for testing multiple variants of the same idea. If you test 20 SMA-crossover variants with different parameters, the laws of probability say at least one will look profitable by pure chance — even if none of them actually are. Bonferroni widens the confidence interval in proportion to the number of variants tested, making false positives much less likely.
Example. A 5% significance threshold across 20 variants becomes 0.25% per variant. The bar is harder to clear, fewer false-positives sneak through.
See also: Confidence interval, Statistical significance
A way to estimate a range of plausible outcomes rather than a single number. Take the actual trades your strategy produced. Randomly resample them (with replacement) 10,000 times. Compute the annualised return of each resampled set. The 2.5th and 97.5th percentiles of that distribution are your 95% confidence interval.
Example. Your backtest returned 8% annualised. The bootstrap CI says the true expected return is between -2% and +18% per year. That range is much more useful than the single 8% number.
See also: Confidence interval, Bonferroni correction
The simplest possible strategy: buy the asset at the start, hold it until the end, sell. Used as the benchmark — your fancy strategy has to beat this, otherwise you might as well just buy and hold.
Why it matters. Buying and holding SPY for the last decade returned about 12%/year. A strategy returning 8%/year over the same period lost 4 percentage points/year to the simplest possible alternative — that's a strategy you shouldn't run.
See also: Annualised return
The fee a broker charges per trade. Most US retail brokers offer zero commission on US equities since 2019, so EdgeAudit defaults to 0% for stocks. Crypto venues typically charge 0.1% per side. Strategies that look profitable before commissions sometimes turn unprofitable once realistic fees are deducted — EdgeAudit applies commissions during the backtest so you see the after-fees result.
A range of plausible values for a statistic, computed from data. A 95% confidence interval means: if you ran this same experiment many times, the true value would fall inside this range 95% of the time. EdgeAudit reports two: the standard 95% interval, and a stricter Bonferroni-corrected one that accounts for testing multiple strategy variants.
See also: Bootstrap CI, Statistical significance
A strategy that triggers when one moving average crosses above (or below) another. Buy signal: short-term average crosses above long-term — the recent trend is now stronger than the long-term trend. Exit signal: short-term crosses back below. EdgeAudit offers three crossover presets via /recipe: Golden Cross, Quick Cross, MACD-style.
See also: Golden Cross, MACD, SMA, EMA, Momentum
The opposite of a Golden Cross. Occurs when the 50-day average price falls below the 200-day average, signalling the upward trend has broken. In the EdgeAudit Golden Cross recipe, a Death Cross is the sell signal.
See also: Golden Cross, Crossover strategy
The largest peak-to-trough loss you'd have suffered while running the strategy. Always shown as a negative number (e.g. -18.7%). A strategy with great returns but a -50% drawdown is psychologically very hard to actually run — most people abandon it during the drawdown and miss the recovery. Always look at drawdown alongside returns.
Example. Your equity goes from £10,000 → £12,000 → £8,500 → £14,000. The max drawdown is the £12,000 → £8,500 fall: -29.2%.
See also: Sharpe ratio, Annualised return
A moving average that weights recent prices more heavily than older ones, so it reacts faster to changes than a Simple Moving Average (SMA). The 12-period and 26-period EMAs are the classic MACD configuration. Useful when you want a moving-average signal that catches turns sooner.
Try it: /recipe company: Apple strategy: MACD-style years: 5
See also: SMA, MACD, Crossover strategy
The most famous pattern in technical analysis. Triggers when a stock's 50-day average price crosses above its 200-day average. The intuition: the 50-day represents the recent trend, the 200-day represents the long-term trend. If recent is above long-term, the stock is in an upward phase. The reverse signal (50-day falls below 200-day) is the Death Cross.
Reality check. The 50/200 crossover is slow. On most single stocks it fires only 2-5 times over 5 years — usually too few trades to draw a statistical conclusion. Works better on broad indices like SPY.
Try it: /recipe company: SPY strategy: Golden Cross years: 10
See also: Death Cross, Crossover strategy, SMA, Momentum
A portion of the historical data set aside and never touched during strategy development. After you've picked your strategy's parameters using the rest of the data, you test it once on the holdout. That single result is the honest estimate of what the strategy would do on data it hasn't seen.
See also: In-sample / out-of-sample, Walk-forward validation
In-sample = the historical data your strategy was built and tuned on. Out-of-sample = data the strategy hasn't seen, used to honestly test performance. The whole point of out-of-sample testing is to avoid overfitting: if a strategy performs well in-sample but poorly out-of-sample, you've fit to noise. EdgeAudit uses walk-forward validation plus a final untouched holdout to enforce out-of-sample testing.
EdgeAudit's verdict when the strategy generated fewer than 30 trades over the test window. With fewer trades, confidence intervals are too wide to be meaningful and win/loss rates aren't distinguishable from coin flips.
What to do. Try a faster strategy on the same asset (MACD or any RSI preset typically fires 10-100× more often than Golden Cross), or extend the test window to 10-20 years.
See also: Verdict, Bootstrap CI
Moving Average Convergence Divergence — a technical indicator developed in the late 1970s by Gerald Appel. The signal at the heart of MACD is whether the 12-period EMA is above or below the 26-period EMA. EdgeAudit's "MACD-style" recipe runs exactly this crossover. Fires more often than the slower Golden Cross — typically 10-30 trades over 5 years on a single stock — so it's good for getting statistical signal.
Try it: /recipe company: AAPL strategy: MACD-style years: 5
See also: EMA, Crossover strategy, Momentum
The bet that sharp price moves tend to reverse rather than continue. Buy when the stock has fallen sharply (the assumption: it's overshot, will bounce). Sell when it's risen sharply. Opposite of momentum strategies. RSI-based strategies are the most common mean-reversion approach.
The bet that recent price moves tend to continue. Buy when the stock is in an upward trend (recent average above long-term average). Sell when the trend breaks. Opposite of mean reversion. Crossover strategies like Golden Cross and MACD are momentum strategies.
When a strategy looks great on the historical data it was tuned on but performs poorly on new data. Usually happens when you try many parameter combinations on the same data and pick the best — you've fit to random noise in that specific history, not a real edge. The single biggest reason retail backtests fail in live trading.
Defences. Bonferroni correction (penalises multiple comparisons), walk-forward validation (tests on data the strategy hasn't seen), and the holdout discipline (test once, never iterate). EdgeAudit applies all three.
See also: Blog: Why most retail backtests lie
The strongest verdict EdgeAudit gives. The strategy had enough trades (30+) for statistical power, AND the lower bound of both the standard 95% confidence interval and the stricter Bonferroni-corrected interval stayed positive. Even after adjusting for the parameter variants typically explored, the result is unlikely to be luck.
A PASS doesn't guarantee future performance — markets change — but it's a meaningful evidence bar that most retail backtests can't clear.
The strategy passed the lighter check — the standard 95% confidence interval excluded zero — but didn't survive the stricter Bonferroni-corrected interval that adjusts for multiple parameter variants. Worth investigating further; not ready to deploy with real capital.
What to do. Try varying the parameters slightly. If a small change wipes out the result, it's probably overfit noise. If similar variants also look promising, the underlying idea has some signal.
The statistical evidence isn't strong enough to call the strategy real. Either the point estimate of returns is negative, or the 95% confidence interval crosses zero — meaning we can't rule out that the apparent edge is luck.
What to do. Discard this exact parameter combination. Don't iterate on it — tweaking until something passes just compounds the multiple-comparison problem. Try a fundamentally different strategy on the same asset.
A 0-100 indicator that measures how strongly a stock has been gaining or losing recently. Below 30 is conventionally called "oversold" (the stock has fallen sharply, may bounce). Above 70 is "overbought" (risen sharply, may correct). RSI mean-reversion strategies buy at oversold levels and sell at overbought levels — the opposite of momentum strategies like Golden Cross.
EdgeAudit has three RSI presets: Classic (30/70 thresholds, textbook), Conservative (25/75, only extremes), Aggressive (35/65, more trades). All use the standard 14-period RSI lookback.
Try it: /recipe company: AAPL strategy: RSI Classic years: 5
See also: Mean reversion
A risk-adjusted return measure: average return divided by the standard deviation of returns. Rule of thumb: above 1.0 is solid, above 2.0 is exceptional, below 0.5 is weak. Higher = more return per unit of volatility.
Be cautious. A single Sharpe number with no confidence interval can be misleading — Sharpes computed on small trade samples are very unstable. EdgeAudit reports the in-sample Sharpe but the verdict layer relies on bootstrap CIs, not Sharpe alone.
See also: Drawdown, Annualised return
The average price over the last N periods, weighted equally. A 50-day SMA on a stock is just the mean of its closing prices over the last 50 trading days. Recomputed each day, so it "moves" as new prices arrive. The slower, smoother counterpart to the EMA.
See also: EMA, Golden Cross, Crossover strategy
The probability that a result you're seeing in your data is real rather than a fluke. By convention, "statistically significant at 5%" means there's less than a 5% chance you'd see this result if there were no real effect. EdgeAudit uses the more conservative idea of a 95% confidence interval excluding zero, with a Bonferroni correction for testing multiple variants — a higher bar than the textbook 5% threshold.
A rule that tells you when to buy and when to sell. That's it. A good strategy has clear, mechanical rules — no "buy when you feel like the market's good".
Examples. "Buy when 50-day SMA is above 200-day SMA, sell when it isn't" (Golden Cross). "Buy when RSI drops below 30, sell when it rises above 70" (RSI mean-reversion). EdgeAudit's recipes (/recipe) are six famous strategies pre-configured.
See also: Backtesting, Crossover strategy
The unique short code that identifies a publicly traded asset. Apple's ticker is AAPL. Tesla's is TSLA. The S&P 500 ETF is SPY. EdgeAudit accepts company names too — type "Apple" in the company field and the bot looks up the ticker automatically.
See also: Try it in the guided builder
EdgeAudit assigns every backtest one of four verdicts based on its statistical evidence:
A way to test a strategy on data it hasn't seen, while respecting the arrow of time. Instead of one train/test split, you create many: train on the first chunk, test on the next, slide the window forward, repeat. Each prediction is made on data that came strictly after the training window that produced it — no peeking at the future.
Why bother? Overfitting defence. A strategy that performs well across five separate test windows is far more credible than one that performs well on a single window. EdgeAudit applies walk-forward by default plus a final untouched holdout.
See also: Blog: Walk-forward vs train/test split
The percentage of your strategy's trades that were profitable. Often the first number new backtesters look at, but it's much less informative than people think — a 60% win rate where the losers are much bigger than the winners is a losing strategy overall. Look at win rate alongside average win/loss size and the Sharpe ratio.
The bot's /learn command surfaces the same explainers inside Discord while you're running backtests. Get the bot installed and try a recipe.