Improving an Investment Strategy

Objective

Every strategy can be made to look better on its own history. Change a setting, re-run, keep the change if the number went up — repeat that enough times and the backtest will improve without the strategy improving at all.

So this case study sets a stricter goal. An improvement here means a change you can explain before you see the result, that makes a difference you can attribute to it, and that still holds on a stretch of history it was not found on. A higher CAGR alone does not qualify.

Everything below is hypothetical and illustrative — a demo strategy and its backtests, used to show the method, not a strategy to copy or a recommendation.

Approach

One change, one reason, one fair comparison

1. Read the evidenceName one specific weakness in the backtest
2. Form a hypothesisWhich rule causes it, and what changing it should do
3. Change one ruleCustomize — and nothing else
4. Compare the two runsOnly that rule should differ
5. Test another periodDoes the difference survive history it wasn't found on?
6. Keep or rejectOn the evidence, not on the headline

Strategy design: the starting point

Midcap Climbers is a Custom strategy built in the Custom Strategy Builder: the fifteen highest-ranked stocks of the Nifty Midcap 150 by 6- and 12-month momentum, ranked on volatility-adjusted returns, rebalanced monthly, with a 20% Trailing Stop Loss and a 0.1% transaction cost on every buy and sell. Its first backtest, from October 2021 to October 2026, used an Exit Rank of 25 — a holding is sold once it slips below 25th place.

The result was respectable — a CAGR of 17.83% — but one figure stood out on the Performance tab: 233 trades in five years, for a portfolio of fifteen. That is a lot of turnover for a monthly strategy, and with a 0.1% cost on every buy and sell, each of those trades costs something real.

That gave a specific hypothesis: the strategy lets go of holdings too early. The rule responsible is Exit Rank. Widening it from 25 to 30 should let a holding drift a little further before it is sold — cutting trades without changing what the strategy buys. What it should not obviously do is change the return much; if it did, that would itself need explaining.

Review of outcome: the full window

The change was made with Customize on the strategy's results page: Exit Rank from 25 to 30, nothing else, then Update & Run Backtest — a Custom backtest, so it spent one Strategy credit. The new run became the strategy's current result, and the original stayed in its Backtest History.

Ticking both runs there and pressing Compare selected gives the fairest comparison the product offers — and the Settings section, which shows only what differs, proves it:

One row. Whatever separates these two results came from Exit Rank.

Exit Rank 25 against 30, October 2021 – October 2026

Exit Rank 25 (original)

• 233 trades
• Max drawdown −25.89%
• Score 75

=

17.83%

CAGR

Exit Rank 30 (changed)

• 212 trades
• Max drawdown −26.02%
• Score 80

=

18.50%

CAGR

On the full window, the hypothesis looks confirmed and then some: fewer trades, as predicted — and a higher CAGR and Score as a bonus. This is exactly the point at which most improvement loops stop. It is too early.

Review of outcome: does it survive another period?

Both versions were tested over the same five years the change was judged on. To see whether the gain belonged to the rule or to that stretch of history, each version was re-run on two halves of the window separately. (Save As made two copies at no cost, and Customize set each copy's Backtest Period to Custom dates — two more Custom backtests per copy.)

Read across the rows and the full-window story comes apart.

• The return gain did not hold. In the first half, the original rule did better — by more than two points a year. In the second half the two are effectively level: 14.61% against 14.70%, with the same drawdown, Score and Sharpe.
• The reduction in trading did hold. Exit Rank 30 traded less in both halves — 78 against 91, and 117 against 127.

So the honest conclusion is narrower than the full window suggested. The change does what the hypothesis said it would: it trades less. The extra CAGR was a property of how those particular five years happened to line up, not of the rule. Kept for the right reason — lower churn, and so lower real-world costs — the change stands. Kept for the CAGR, it would not.

A second opinion from an Experiment

On plans that include it, an Experiment can vary several settings at once. One run on Midcap Climbers varied its ROC windows and holding count:

Its useful output is not the configuration that came first but the pattern: the top three all held ten stocks, whichever ROC window they used. That is a new hypothesis — a tighter portfolio suits this universe — and it deserves exactly the treatment above: one change, one fair comparison, then a period it was not ranked on. An Experiment proposes; it does not prove.

Trade-offs

When to stop tweaking

At that point the better next step is not another backtest. It is a simulation — a paper portfolio run with virtual money — where the strategy meets conditions no backtest has seen. Historical results, however carefully tested, do not guarantee future performance, and what you do with a strategy remains your decision.

Lessons learned