Backtests & side-by-side compare
The Strategy page is the lab: run a backtest of your active strategy version (a simulation against history), read its equity curve (the chart of the strategy’s value over time) against the S&P 500 (tracked via SPY, a fund that follows the index), inspect walk-forward folds, and, once two or more versions have been backtested, compare them side by side.
Running a backtest
The header button queues a run on demand (there’s no schedule; backtests run when you ask, with a short cooldown, about five minutes, between runs, since each one pulls years of history for your whole universe). Results land in three cards: the equity curve (your strategy vs the S&P 500 — both lines are named in the key beside the chart’s summary row, the 1W · 1M · 3M · YTD · 1Y · ALL buttons zoom the tested window, and the row under the chart shows the tested window, worst peak-to-bottom fall, and rebalance count), the metrics table (Sharpe and Sortino, return earned per unit of risk taken; max drawdown, the worst peak-to-bottom fall; CAGR, average yearly growth; and more), and the walk-forward folds. If a run can’t produce a result — your universe is empty, or too few symbols have enough price history — SleeveFolio sends you a notification instead of leaving the Strategy Lab looking like it never ran.
One caveat: the backtest ranks on price-based factors only: momentum and low volatility. The value and quality factors the live engine uses are left out, because the free data we have for them is a current snapshot with no point-in-time history, and replaying today’s numbers against the past would be cheating (look-ahead). So a version change that only re-weights value or quality will look the same in the backtest even though it changes live signals.
Why walk-forward folds
A single full-period backtest can flatter a strategy that only worked in one kind of market (one long rally, say). Walk-forward testing splits history into time-ordered slices (folds): within each fold the strategy re-ranks your holdings every trading day— the same daily cadence the live engine runs on — using only the prices up to each day (the trailing lookback window), and the fold scores how that daily-rebalanced book actually did over its forward window. The simulated book is traded the way live trading works — it holds a position until a sell signal (or your strategy’s stop-loss), sizes each buy within the per-position and sleeve budgets, trims winners at your take-profit, backs off buying after a steep drawdown, and only ever deploys cash it actually has — so the curve carries real holding inertia and cash drag, not a frictionless snap to target. Because the ranking only ever sees prices from before the day it acts on, every fold’s result is genuinely out-of-sample, earned on data the ranking hadn’t seen. Spread across folds matters as much as the average: a strategy that wins on aggregate but dies in two folds is telling you something. (The tilt has no parameters to fit, so the columns are labeled Lookback and Hold, not train/test.)
Side-by-side comparison
After your first backtest, the Version comparison card explains how to unlock the comparison: tune your strategy in Settings → Strategies, save the new version, and backtest it. With two or more versions backtested, the card lines their metrics up version-vs-version: the honest way to decide whether yesterday’s tweak actually helped before you let it trade.