The transition from "ran a backtest" to "actually validated a strategy" is the single most common gap in retail quant research. This cookbook walks through the steps in order — what to do, what to look at, and what to do next.
Step 1 — Run the Initial Backtest
Specify the strategy in advance: indicators, parameters, rules, risk management. Run the backtest on the full history. Look at:
- Sharpe. First read. If < 1.0, the strategy isn't worth proceeding with. If > 3.0, suspicious — investigate before celebrating.
- Trade count. < 30 is too few; > 1000 is fine if costs are accounted for.
- Equity curve shape. Smooth = good (or curve-fit). Lumpy = real but harder to trade.
- Max DD and DD duration. Can you live with these? If not, no further validation is needed — it's the wrong strategy for you.
Step 2 — Robustness Sweep
Pick the 1–2 most uncertain parameters. Sweep over reasonable ranges. Look at the heatmap:
- Stable plateau: a contiguous green region. Pick somewhere in its interior — never the absolute best cell.
- Isolated peaks: warning. The "best" parameters are probably noise.
- DSR > 0.95: survives multiple-testing. Confidence-builder.
- DSR < 0.80: red flag. Most of the apparent edge is search bias.
If no plateau exists, the strategy is too sensitive to parameters — go back to the drawing board, simplify, or accept that this idea has no edge.
Step 3 — Pick the Plateau Center
Round to clean parameter values inside the plateau. Document the chosen parameters. Lock them.
Step 4 — Walk-Forward
Set up walk-forward with 5–10 folds and your chosen parameter constraints. Choose anchored mode if you trust the regime is stable; rolling if you suspect drift.
Read the output:
- Composite OOS Sharpe vs IS Sharpe. Decay ratio (OOS / IS) should be > 0.5; ideally > 0.7.
- Per-fold equity. All folds positive? Or one disastrous fold pulling the average up?
- Parameter stability across folds. Did the optimizer converge to similar parameters in each fold? Drifting parameters → fitting noise.
- Composite max DD. Often the most honest forward-looking DD estimate.
If walk-forward Sharpe is > 60% of in-sample Sharpe with stable per-fold results, you have a tradable strategy. If walk-forward Sharpe is < 30% of IS Sharpe, the strategy is mostly curve-fit. If 30–60%, the edge is real but partial — proceed with smaller size.
Step 5 — Reveal Out-of-Sample (Single Reveal)
If walk-forward looks good, you can also do a single OOS reveal on the platform's reserved OOS window. This is a one-shot test — once you see it, you can't unsee it. The result should broadly agree with walk-forward.
Step 6 — Save and Move to Paper Trading
Save the config to your Trading Configs library. Push to paper trading. Now you wait — the next 3–6 months of live data is the truest validation. Compare paper P&L to backtest expectation periodically; if it diverges materially, run a post-mortem.
What to Avoid Between Steps
- Don't modify the strategy after seeing OOS. Once revealed, OOS is contaminated.
- Don't rerun the sweep with different ranges if the first sweep was disappointing. Each rerun adds search bias.
- Don't cherry-pick which folds to keep. All folds count.
- Don't skip steps because earlier steps looked good. A 4.0 Sharpe in step 1 makes step 2 (robustness) more important, not less.
The Bottom Line
Backtest → Sweep → Walk-Forward → Reveal → Paper. Each step takes more time and adds more honesty. Most strategies die at one of the steps; that's the validation working. The strategies that survive all the way to paper trading are the ones worth real capital.
Further Reading
Foundational papers
- Pardo, R. (2008). The Evaluation and Optimization of Trading Strategies (2nd ed.). Wiley.
- Bailey, D. H., Borwein, J. M., López de Prado, M. & Zhu, Q. J. (2014). Pseudo-Mathematics and Financial Charlatanism: The Effects of Backtest Overfitting on Out-of-Sample Performance. Notices of the AMS, 61(5), 458–471.
Textbook references
- Pardo, R. (2008). The Evaluation and Optimization of Trading Strategies (2nd ed.). Wiley.
- López de Prado, M. (2018). Advances in Financial Machine Learning. Wiley.
Related QuanterLab articles
- Walk-Forward Validation
- Robustness Sweeps
- Deflated Sharpe Ratio
- Cookbook: Reading the Robustness Heatmap
- Cookbook: Interpreting Walk-Forward Results
Try it in QuanterLab
Pick a strategy you have already saved. Walk through this cookbook on it: run the backtest, sweep one parameter, walk-forward, and compare the verdict to your prior intuition. Most strategies do not survive the journey — that is the validation working.