How should a backtest weight stock positions?
A backtest should start with the simplest weighting rule that answers its research question. For a stock-selection thesis, equal weighting is usually the clean baseline: every held company has the same influence, so market size doesn't bury the signal. Then challenge that baseline with plausible alternatives before giving the thesis credit.
Position weighting is the rule that divides the hypothetical portfolio among companies that qualify. It can change the result even when the screen, dates, entries, and exits stay fixed. If a thesis works only under one narrow weighting rule, the portfolio policy is carrying part of the finding.
Why is equal weighting a useful baseline?
Equal weighting asks a clean question: what happened when every active qualifier received the same influence? The arithmetic is easy to inspect:
weight_i = 1 / active_positionsThat rule keeps a very large company from dominating the path because it has a large market value. It also stops a thesis score from becoming an allocation rule before you've shown that higher scores deserve more capital. A passing company is a passing company.
Quantery's naive portfolio uses equal-weight positions under the saved entry and exit rules. The word naive matters. This is a consistent research model. It doesn't reconstruct anyone's brokerage account. The guide to reading a backtest report explains why its curve belongs beside the event study as a separate view.
Equal weighting still makes choices. It gives smaller companies more influence than a market-cap-weighted portfolio would. Rebalancing it trims positions that have grown and adds to positions that have lagged. When only a few companies are active, each one matters a great deal. None of that is a flaw hidden in the calculation. It is the policy being tested.
Use equal weight because it is plain. Don't mistake it for neutral. There is no neutral allocation once several companies compete for the same hypothetical dollar.
How can position weights change a backtest result?
The same qualification events can produce different curves through several mechanisms.
Company concentration. A weighting rule can let one holding control much of the portfolio. That holding's path then looks like evidence for the whole thesis. Equal weighting limits concentration at a rebalance, but a winner can grow into a much larger share before the next reset. Outlier-dependence tests deal with the same problem from the contribution side.
Shared exposure. Equal company weights don't guarantee diversified economic exposure. Several holdings may depend on the same commodity, customer group, or financing environment. FINRA defines concentration risk as the chance of magnified losses when too much of a portfolio sits in one investment or market segment, and it warns that correlated holdings can hide that concentration. A basket of different tickers can still be one bet.
Rebalancing pressure. Returning positions to target weights creates trades. A rule that resets frequently can harvest reversals in the historical sample, or repeatedly add to deteriorating companies. The gross curve won't tell you which story applies. Quantery's hypothetical backtest returns exclude trading costs, so any benefit that arrives with more turnover needs a cost sensitivity check. The trading-cost stress test shows how to frame that check.
Signal intensity. Score weighting gives more influence to companies with higher thesis scores. That sounds natural, but a score is often ordinal: it says one company cleared more bands. It doesn't specify an expected return difference. Weighting by it can count the same signal twice, first in selection and again in allocation.
Capacity and liquidity. Market-cap or liquidity weighting pushes more hypothetical capital toward companies that may be easier to trade. It also changes the research claim. The result now reflects the size or trading-volume rule along with the accounting thesis.
Which weighting rule matches the thesis?
Write the claim before choosing the allocation. The right baseline follows from what the thesis says should matter.
If the claim is that a fundamental pattern identifies unusually priced companies across an eligible universe, equal weight is a sensible first test. It lets each selection speak at the same volume.
If the claim says the score itself ranks future outcomes, don't jump straight to score weighting. First test whether stricter score bands show an ordered return pattern. A score-ladder test asks that question without mixing it with portfolio allocation. Only a score that shows useful ordering has earned a weighting experiment.
If the claim concerns an implementable policy for a larger pool of capital, liquidity belongs in the test. A dollar-volume floor can remove companies the policy couldn't trade reasonably. Weighting the remainder by liquidity is a further choice, and it may smother the very small-company effect the original thesis found. Report that change in the claim.
Sparse theses need a cash rule. When few companies qualify, does the unused allocation stay in cash, spread across the available names, or move to a benchmark? Spreading it across the names makes the portfolio more concentrated precisely when the signal is scarce. Holding cash makes signal availability part of the result. Neither answer can be filled in after seeing which curve looks better.
Sector limits require the same clarity. A cap can answer an implementation question, but it can also remove a real sector-specific mechanism. Run the unrestricted version first, then the constrained version. If the conclusion changes, use the sector-bet test before deciding which version tells the more useful story.
How do you run a weighting sensitivity test?
Freeze the thesis version, universe, date range, benchmark, reporting lag, rebalance cadence, entry rule, and exit rule. Preserve the same qualification events. Recompute only the allocation policy.
Start with equal weight. Add the nearest plausible challenger, chosen from the claim before you see the finished chart. That might be a per-company cap with residual cash, a liquidity-aware rule, or a score-aware rule whose score ordering has already been tested. You don't need a gallery of clever formulas. You need a disagreement you can interpret.
Define the mechanics completely before calculating:
- Say when weights are set and when they reset.
- Say how a new entrant receives capital.
- Say what happens to weight released by an exit.
- Say whether an unfilled allocation remains cash.
- Say how company and sector limits interact.
Those details prevent a label such as "capped equal weight" from hiding several different simulations. A cap with leftover cash answers a different question from a cap whose excess is spread across the remaining holdings.
Quantery currently gives you the naive equal-weight portfolio as the reproducible baseline. It doesn't produce custom-weight portfolio variants in the report. If weighting is central to the research question, reconstruct the baseline outside the app before changing it. Match the saved entries, exits, rebalance dates, dividends, delisting treatment, and aggregate path. If your reconstruction can't reproduce the baseline, its alternative curve is testing a different machine.
Keep the comparison in a backtest research log. Record the weighting rule, the reason it fits the claim, and the outcome that would weaken your conclusion. This stops allocation from becoming another set of dials turned until the curve improves.
What should survive across weighting rules?
Don't demand identical returns. Different weights must produce different paths. Ask whether the part of the result you care about survives.
Start with direction against the same benchmark. If every plausible rule points the same way while the magnitude moves, the stock-selection claim is stronger than the headline return. Keep the direction and treat the return as policy-dependent.
Then read maximum drawdown. A result that preserves return by loading heavily into one volatile holding may be a different policy in every practical sense. Inspect the largest position through time, sector exposure, and the dates when the paths separate. The divergence usually has a name and a period attached to it.
Compare turnover next. A weighting variant that improves the gross curve through constant resets hasn't established an advantage after spreads, slippage, or taxes. Backtested returns remain hypothetical and exclude those costs. A more active rule has to clear a higher evidentiary bar.
Finally, compare the event study with every portfolio variant. The event study asks what followed fresh qualifications without requiring a portfolio weighting rule. If its median and hit-rate pattern remains stable while portfolio curves disagree, the signal may contain information but the allocation policy is unresolved. If the event study is weak and only a concentrated portfolio looks strong, investigate the holdings before crediting the gate.
A useful conclusion names both layers: the qualification rule produced a certain event pattern, and a stated weighting policy turned those events into a particular portfolio path. Keep those claims separate. Change the weights and rerun the policy. Change the gate and rerun the signal. The rules are yours, including the rule that decides how much influence each survivor gets.
Research tooling, not investment advice. Nothing here is a recommendation to buy, sell, or hold any security. Screens, scores, and backtests are informational only; backtested results are hypothetical, exclude costs such as commissions and slippage, and do not guarantee future results. Verify against primary filings and make your own decisions.
Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.
← All articles