1. Calculators
  2. Quant & Algorithmic Trading

Walk-Forward Efficiency Calculator

Check how much of a strategy's optimized performance survives on data it has not seen. Enter the in-sample and out-of-sample returns of each walk-forward window to get the walk-forward efficiency, the share of profitable windows and how consistent the results are.

Updated Oct 16, 2026Free · no sign-upFormulas tested

Your inputs

Enter one row per walk-forward window: the annualized return of the optimized parameters in the in-sample period and in the out-of-sample period that followed.

Walk-forward windows
WindowIn-sample return (annualized) %Out-of-sample return (annualized) %IS length (months)OOS length (months)
%
Rule of thumb: about 50–60% is often cited as acceptable. Windows at or above it count as consistent.

Results

Walk-forward efficiency40%Average out-of-sample return ÷ average in-sample return, weighted by window length. Below your threshold: much of the in-sample return did not survive
Average in-sample return
+25.59%
Average out-of-sample return
+10.26%
Windows profitable out of sample
88%
7 of 8
Windows with efficiency of 50% or more
63%
5 of 8: consistency across windows
Median window efficiency
52%
Average of window efficiencies
41%
Weakest window
-23%
Window 4

Out of sample, the strategy kept 40% of its in-sample return: 10.26% a year against 25.59% when the parameters were fitted. 1 of 8 windows lost money out of sample.

In-sample vs out-of-sample return by window (annualized)

In-sampleOut-of-sample
-10%0%10%20%30%40%Window 1Window 2Window 3Window 4Window 5Window 6Window 7Window 8

Each pair is one optimization: the blue bar is the return of the best parameters on the data they were fitted to, the amber bar their return on the next, unseen period.

Efficiency by window

WindowIn-sampleOut-of-sampleEfficiencyReading
Window 1+24.5%+14.2%58%Held up
Window 2+31.0%+9.8%32%Faded
Window 3+19.8%+15.1%76%Held up
Window 4+27.3%-6.4%-23%Loss out of sample
Window 5+22.6%+12.9%57%Held up
Window 6+35.1%+18.3%52%Held up
Window 7+18.4%+4.7%26%Faded
Window 8+26.0%+13.5%52%Held up

Results are estimates from the numbers you enter. They are not financial, tax or legal advice.

How to use the walk-forward efficiency calculator

  1. Run a walk-forward analysis in your backtesting software: optimize the parameters on an in-sample window, trade the best parameters on the out-of-sample window that follows, then roll both windows forward and repeat.
  2. Enter one row per window: a label, the annualized in-sample return of the chosen parameters, and their annualized out-of-sample return. The example has eight windows of a fictional strategy.
  3. Add the window lengths if you want the averages weighted by time, for example 24 months in sample and 6 months out of sample. Leave them empty to weight every window equally.
  4. Set the acceptable efficiency, 50% by default. Windows at or above it count as consistent.

The formula

WFE = ROOS ÷ RISROOS is the annualized out-of-sample return and RIS the annualized in-sample return of the same parameter set. It is defined only when RIS is positive.
Overall WFE = average ROOS ÷ average RISEach average is weighted by window length when every row has one, otherwise equally. The calculator also shows the median and the simple average of the window efficiencies.

Annualizing matters because the two windows rarely have the same length. Comparing 18 months of in-sample profit with 6 months of out-of-sample profit would make almost any strategy look degraded.

A worked example

The eight windows each optimize on 24 months and trade the next 6. In sample, the chosen parameters returned an average of 25.59% a year; out of sample, 10.26%. The walk-forward efficiency is 40%, below the 50% rule of thumb.

The detail tells a more nuanced story. The median window kept 52% of its in-sample return, 88% of windows made money out of sample, and 63% cleared the 50% bar. One window, Window 4, lost money: its efficiency of -23% drags the overall figure down. Had it earned 8% instead, the overall efficiency would have been 47%.

How to read the result

Overall versus median. The overall WFE compares average returns, so one disastrous window pulls it down hard. The median shows what a typical window did. When the two differ a lot, look at the window table: is the weak window a one-off event or the start of a trend?

Consistency. The share of windows above your threshold, and the share profitable out of sample, show whether performance is spread evenly or concentrated in a few lucky periods. A strategy that clears the bar in most windows is more trustworthy than one with the same average built on two outstanding windows. Raising the threshold to 60% in the example leaves only 13% of windows passing.

Why efficiency is below 100%. Optimization picks the parameter set with the best in-sample result, and part of that result is noise that happens to fit the period. That part does not repeat, so the in-sample return overstates what the parameters will earn next. Markets also change between windows. Some loss of performance is expected; the question is how much.

Absolute results. Efficiency is a ratio, so it can flatter a weak strategy: 4% in sample and 3% out of sample is a 75% efficiency on a return that may not cover costs. Judge the out-of-sample returns on their own as well, ideally as one stitched-together equity curve.

Walk-forward analysis and overfitting

Walk-forward testing guards against fitting one period, but it can be overfit itself. If you try many window lengths, re-optimization schedules and parameter ranges, and keep the combination with the best walk-forward result, the out-of-sample periods are no longer truly unseen. Count those choices as trials. The probability of backtest overfitting estimates how often the in-sample winner falls below the median out of sample, and the deflated Sharpe ratio adjusts a Sharpe ratio for the number of trials.

It also pays to look beyond returns: compare the drawdowns, win rates and trade counts in and out of sample. A strategy whose out-of-sample trades look nothing like its in-sample trades is behaving differently, whatever its efficiency.

What this calculator does not include

  • Risk. WFE compares returns, not risk-adjusted returns. Two windows with the same return can carry very different drawdowns.
  • Compounding across windows. Averages are arithmetic means of annualized returns, weighted by length if given; they are not the compound return of the stitched out-of-sample equity curve.
  • Negative in-sample windows. Efficiency is undefined when the optimized in-sample return is zero or negative; those windows are excluded from the window-efficiency statistics but kept in the averages.
  • Statistical significance. A handful of windows cannot prove skill. Use the out-of-sample returns with the Sharpe ratio calculator or a Monte Carlo trading simulation for a sense of the range.
  • Costs. Enter returns after commissions, spreads and slippage, in and out of sample alike.

Frequently asked questions

What is walk-forward efficiency?
Walk-forward efficiency (WFE) is the annualized return a strategy earned out of sample divided by the annualized return it earned in sample, on the data its parameters were optimized on. A WFE of 60% means the strategy kept 60% of its optimized performance when traded on unseen data. Robert Pardo popularized the measure as part of walk-forward analysis.
What is a good walk-forward efficiency?
A rule of thumb often cited in the trading-system literature is that about 50% to 60% or more is acceptable, and lower values point to overfitting or a strategy that does not adapt well. It is a convention, not a statistical test. Check the out-of-sample return in absolute terms too: a high efficiency on a weak in-sample return is still a weak strategy.
Why are the returns annualized?
Because in-sample and out-of-sample windows usually have different lengths. A 24-month optimization window and a 6-month trading window cannot be compared on total return; annualizing both puts them on the same footing.
Can walk-forward efficiency be above 100%?
Yes, when the out-of-sample period happens to be more favorable than the in-sample period. A single window above 100% is not unusual, but an average well above 100% across many windows is suspicious: check for look-ahead bias, mismatched windows or out-of-sample periods that overlap the optimization data.
How many walk-forward windows do I need?
More windows give more out-of-sample evidence. With only a few, one good or bad window decides the result. In the example, 8 windows of six months each add up to 4.0 years of out-of-sample trading, a reasonable minimum for a strategy that trades often.

Related calculators

All calculators →

This calculator is for education and planning. It uses the formulas described above and the inputs you provide; real results depend on fees, taxes, timing and other details that a general tool cannot know. PromtFinance does not give personal financial advice. Check important decisions with your provider or a qualified professional.