Reading Walk-Forward Results

How to interpret every component of the walk-forward results page -- WFE score, summary cards, IS vs OOS chart, equity curve, parameter stability, and recommended parameters.

Reading Walk-Forward Results

After a walk-forward analysis completes, Simula presents the results across several components: a WFE hero section, summary cards, charts, parameter stability analysis, and a per-window breakdown. This article walks through each one and explains what to look for.

WFE Score

The first thing you see is the aggregate Walk-Forward Efficiency score, displayed prominently with its qualitative tier. This is your headline result -- did the strategy retain its training performance on unseen data?

If the WFE is below 50%, the analysis is telling you the optimized parameters did not generalize. This does not necessarily mean the strategy is worthless, but it does mean the specific parameter values found by the optimizer are not reliable for forward use.

Summary Cards

Below the WFE score, six summary cards provide a quick overview of the analysis.

Windows

How many optimize-then-validate cycles were run. More windows means more out-of-sample data points and a more reliable result.

Profitable OOS %

The percentage of out-of-sample windows that were profitable. A high percentage indicates the strategy finds working parameters consistently, not just occasionally. 75% or higher is a good signal. Below 50% means the strategy lost money on most unseen periods.

Total OOS Return

The compounded return across all out-of-sample segments. This is what the strategy would have earned if you had run walk-forward live, re-optimizing at each step and trading the OOS periods with the latest parameters.

Avg OOS Return

The average return per out-of-sample window. This gives you a sense of per-window magnitude -- are the profitable windows returning 2% or 20%?

OOS Trades

The total number of trades across all out-of-sample windows. This matters for statistical significance -- more trades mean the results are less likely to be driven by luck.

Max Drawdown

The largest peak-to-trough decline in the combined OOS equity curve. This tells you the worst pain the strategy would have inflicted if you had followed the walk-forward parameters through time.

IS vs OOS Bar Chart

This chart shows the in-sample and out-of-sample annualized return side by side for each window. It is the most direct way to see whether the strategy retains its training performance across time.

What Healthy Degradation Looks Like

Some performance drop from IS to OOS is normal and expected. The optimizer had the advantage of seeing the IS data. A consistent moderate drop -- say, IS returns of 15-20% and OOS returns of 8-12% across all windows -- is a healthy pattern. It means the strategy has a real edge that degrades gracefully on unseen data.

Red Flags

Watch for windows where IS performance is high but OOS performance is negative. One or two such windows across many is tolerable -- markets have bad periods. But if most windows show profitable IS and negative OOS, the optimizer is overfitting to each training period and the parameters are not generalizing.

Combined Equity Curve

The equity curve stitches together the out-of-sample segments from all windows into a single continuous performance line. It shows what the strategy would have looked like if you had run it live with walk-forward re-optimization at each step.

Look for an upward-trending curve with controlled drawdowns. The curve will not be as smooth as a single optimized backtest because each segment uses different parameters tuned to different training periods. That unevenness is the point -- it is showing you realistic performance, not optimized-in-hindsight performance.

Vertical markers on the chart show where each window boundary falls. This helps you see which time periods contributed to gains and which caused drawdowns.

Rolling WFE

The rolling WFE chart recalculates WFE while progressively dropping the earliest windows. The first point uses all completed windows, the next drops the first window, then drops the first two, and so on -- stopping when only three windows remain. This reveals whether the aggregate WFE is being pulled up or down by older data.

For example, with 6 completed windows: the first point computes WFE from W1-W6, the next from W2-W6, then W3-W6, and finally W4-W6. That gives you four data points showing how WFE shifts as older windows are removed.

An improving trend (WFE rising as early windows are dropped) suggests those early windows were dragging the average down -- the strategy's edge is stronger in recent data. A deteriorating trend means early windows were inflating the average -- the edge may be fading. Even if the aggregate WFE looks acceptable, a clear downward trend should give you pause before deploying the strategy.

Rolling WFE requires 5+ windows and rolling mode only

The rolling WFE chart only appears for rolling walk-forward runs with at least 5 completed windows. It is not shown for anchored runs because the growing training window size confounds the signal -- you cannot tell whether improving WFE reflects a genuine regime trend or simply better optimization from more training data.

Parameter Stability & Recommended Parameters

The parameter stability table shows whether the optimizer chose similar parameter values across windows or wildly different ones. Each parameter is scored by its coefficient of variation (CV) and assigned a stability tier from Very Stable to Unstable. The table also shows the last window's values and the mean across all windows, with buttons to apply either set directly to the strategy builder.

For a full explanation of CV, the stability tiers, how to read the parameter trajectory chart, and which parameter set to use going forward, see the dedicated Parameter Stability article.

Per-Window Drill-Down

Each window can be expanded to show its full details: the parameter values chosen by the optimizer, the equity curve for that OOS segment, individual trades, and the full set of performance metrics. Use the drill-down to investigate windows that stand out -- either unusually strong or unusually weak -- and understand what happened during that period.

When to Re-Run Walk-Forward

Walk-forward results are not permanent. Re-run the analysis when:

Market conditions change significantly. A major regime shift -- a new interest rate cycle, a liquidity crisis, a structural change in the asset -- can invalidate even robust results. If the market environment has changed meaningfully since the last analysis, re-validate.

Periodically as a routine check. Even without obvious regime changes, running walk-forward every few months on updated data helps confirm the strategy's edge is still present.

After modifying the strategy logic. Any change to the entry/exit conditions, indicators, or structure of the strategy invalidates previous walk-forward results. The new logic needs its own validation.

Red Flags -- When to Abandon a Strategy

Walk-forward analysis is as valuable for rejecting bad strategies as it is for confirming good ones. Consider abandoning a strategy when:

WFE is consistently below 35% across many windows. The optimizer is finding patterns that do not generalize. The strategy logic may not capture a real market behavior.

Parameters are unstable with poor OOS performance. If the optimizer picks wildly different values each window and none of them work out of sample, the strategy has no stable edge to optimize around.

The combined equity curve trends downward. If following the walk-forward parameters through time would have lost money overall, the strategy is not viable regardless of what individual metrics say.

Recent windows are uniformly worse than earlier ones. A strategy whose OOS performance was good three years ago but has been declining steadily may be facing a structural market change that is eroding its edge.

Do not over-tune the walk-forward configuration itself

If you run walk-forward, see a failure, change the window length, run it again, see another failure, change the IS/OOS ratio, and keep adjusting until it passes -- you are overfitting the walk-forward configuration to produce the result you want. Choose reasonable defaults (rolling windows, 3:1 or 4:1 IS/OOS ratio, enough windows for meaningful coverage) and accept the result. The point of walk-forward is to get an honest answer, not a favorable one.

Where to Go Next

Walk-Forward Efficiency -- deep dive into the WFE metric, the 50% threshold, and what matters more than the number itself.

Walk-Forward Analysis -- the concept and method behind the results.

Reading Optimization Results -- how to interpret the parameter optimization results that feed into walk-forward.

Overfitting & Curve Fitting -- the problem that walk-forward analysis is designed to detect.