Methodology
Backtest Methodology
Every strategy on the LisaQuant Desk is run through the same replay engine before it can ever reach the live room. This page explains exactly how that engine works, what the numbers mean, and the guardrails a strategy must clear.
1. The replay process
Point-in-time data only
We replay every strategy against historical daily OHLC bars from a price feed that did not include future information. The engine looks at the signal bar as it would have appeared at that day's close, not after the fact. No adjusted hindsight, no snooping, no peeking at tomorrow.
Execution at the close
All entries and exits are assumed to take place at the closing print of the signal bar. This matches the realistic workflow where the nightly scan identifies a trigger and a member has the next session to act. Friction from slippage, commissions, or spreads is not modeled.
No cherry-picking
The backtest includes every valid signal that the rules produce over the full sample period. We do not remove losing trades, optimize parameters to a specific stock, or rerun the engine with different rules to make a curve look better.
Universe is fixed
Each strategy trades a predefined universe: S&P 500 large caps, liquid ETFs, or a specific index set. The universe is not widened after the fact to include symbols that would have produced the best results.
2. What the metrics mean
R (risk unit)
The fixed amount of capital risked on a trade. If a strategy risks $100 per trade and a closed trade gains $240, that trade returned +2.4 R. R normalizes everything so you can compare strategies regardless of account size.
Win rate
Percentage of closed trades that finished with a positive R result. High win rate alone is misleading; it must be paired with the average winner and loser.
Payoff ratio
Average winner divided by average loser in R. A 2.0 payoff ratio means the typical win is twice as large as the typical loss.
Expectancy in R
The expected return per trade: (win rate × average win) + (loss rate × average loss). Positive expectancy means the strategy should grow capital over many trades. Negative expectancy means it will not.
Max drawdown
The largest peak-to-trough decline in the backtest equity curve, measured from the highest prior equity value. This is the worst losing streak the replay produced.
Average hold
The average number of sessions from entry to exit, including winners and losers. This tells you whether a strategy behaves like a quick swing trade or a longer investment.
95% confidence interval
A statistical range around the average per-trade result. We require the 95% confidence interval to exclude zero before a strategy is promoted. If the interval includes zero, the strategy's edge is not statistically distinguishable from luck.
Sample size
The number of closed trades in the backtest. The guardrail requires a large sample so that the win rate, expectancy, and confidence interval are not driven by a handful of outliers.
3. Hold-window assumptions
Held overnight up to three months
A swing trade is any position that is not closed the same day it is opened and is expected to be held for up to three months. The actual exit is governed by the strategy's rules: an initial stop, a target, a time stop, or a trailing stop after a minimum holding period. We do not hold losing swing trades past their hard stop just because the calendar has not hit three months.
Held three months and beyond
An investment trade is designed to capture multi-month or multi-quarter trends. The backtest still applies a structured exit rule, but the expected hold window is materially longer than a swing. The strategy is tested over a window that includes at least one full multi-year cycle to see how it behaves across bull, bear, and choppy markets.
4. The LisaQuant Guardrails
No strategy or trade reaches the live LisaQuant Trading Desk until it clears all of the backtested AI reverse-engineered algorithms for full ROI optimization. The promotion gate includes:
- Positive expectancy measured in R
- 95% confidence interval that excludes zero
- Strategies used backtested over a large sample size
- Reported win rate, payoff, max drawdown, and average hold
5. Important limitations
All backtest results are hypothetical. They assume execution at historical closing prices, ignore slippage, commissions, taxes, and market impact, and may not reflect the liquidity available on the actual day of a signal. A strategy that performed well in a historical replay is not guaranteed to perform the same way in live markets. Past or simulated results do not guarantee future outcomes.