If you are building a scanner, the first instinct is to poll faster. A five-second refresh feels obviously better than a one-minute refresh, and the cost is real — more requests, more rate-limit pressure, more moving parts.
We measured whether it actually catches more. It does not, at least not in a way that would justify the cost.
1. What we measured
A "catch" here means: the symbol was already on the watchlist five minutes before its move. Five minutes is the lead time a strategy needs to have any chance of acting.
The watchlist is rebuilt every N seconds from the market at that moment — turnover leaders unioned with percentage-change leaders, held as a 30-minute rolling union. K is the size of the list.
We ran it in both directions from the 60-second baseline, because a backtest built on minute bars already assumes 60-second refresh: faster on trade-level data, slower on minute bars.
2. Faster than 60 seconds (trade-level data, 60 sessions)
| Refresh | K=300 | K=500 | K=750 |
|---|
| 5s | 78.3% | 90.6% | 94.6% |
| 10s | 78.0% | 90.4% | 94.6% |
| 15s | 77.9% | 90.3% | 94.5% |
| 30s | 77.2% | 89.9% | 94.2% |
| 60s | 76.4% | 89.4% | 93.8% |
Going from 60 seconds to 5 seconds bought 1.2 percentage points at K=500. Twelve times the polling rate for just over one point.
3. Slower than 60 seconds (minute bars, 501 sessions)
| Refresh | K=300 | K=500 | K=750 |
|---|
| 60s | 78.8% | 90.6% | 94.0% |
| 120s | 77.8% | 90.0% | 93.6% |
| 180s | 76.9% | 89.4% | 93.2% |
| 300s | 75.6% | 88.4% | 92.6% |
This is the direction that would have worried us — a broker that only lets you sweep the market every few minutes. Stretching the refresh from 60 to 300 seconds cost 2.2 points.
Across the whole span, 5 seconds to 300 seconds, the catch rate stayed between 88.4% and 90.6%.
4. Why the interval matters so little
The rolling 30-minute union absorbs the delay. A symbol that enters the leaderboard is still in the union several refreshes later, so missing one snapshot rarely means missing the symbol. What decides the catch rate is how many names you are willing to watch, not how often you look.
That shows up clearly in the size dimension. At the same 5-minute lead time:
| Watchlist size | Catch rate |
|---|
| 208 | 62.5% |
| 280 | 72.4% |
| 358 | 80.4% |
| 561 | 90.6% |
Widening the list moved the number by 28 points. Changing the refresh interval by a factor of 60 moved it by 2.
5. Lead time is the other real variable
| Lead | 0 min | 5 min | 10 min |
|---|
| Catch rate | 97.9% | 90.6% | 86.9% |
Almost everything is on the list by the time the move happens; the question is how much warning you get. Across four half-year sub-periods the 5-minute figure stayed between 89.5% and 91.7%, so this is not an artefact of one regime.
Conclusion
If your scanner is missing movers, polling faster is unlikely to be the fix. Between 5 seconds and 5 minutes the difference was about two percentage points. How many symbols you watch, and how much lead time you demand, moved the result an order of magnitude more.
