Trading Platform Research

Home / Research examples / Waiting three days after a large drop

Did waiting three days after a large drop help?

Plenty of traders say you should let a falling stock settle before buying. This test compares buying at the close of the drop against buying three trading days later.

Documented

The company says so in its own docs. The source is linked and dated.

Observed

I ran it and this is what happened. Inputs and steps are shown.

Opinion

My judgment. You may weigh the same facts differently.

The test

Question
After a stock falls sharply in one day, compare buying at that close against buying three trading days later.
Inputs
Event: an S&P 500 stock closes down 10% or more in one day, using membership as of the event date. Entries: that close, or the close three trading days later. Holding period: 20 trading days from each entry. Events from January 1, 2020 through December 31, 2025.
Method
  1. Give each platform the same event definition once, in plain English, through its AI assistant.
  2. Measure the return of both entry rules over the same 20-day holding period.
  3. Record the number of events found and the average and median return of each entry.
Result
Observed
NexusTradeQuantConnectComposer
Tested without writing codeYesYes, the assistant wrote the PythonYes
Question to answer4 min 27 sMessage sent 03:28:12 UTC; the answer finished at 03:32:39 UTC.18 min 57 sFrom QuantConnect's own records: the run started at 16:57:44 UTC and its final answer is stamped 17:16:41 UTC.43 sMessage sent 23:22:26 UTC; the answer stopped growing at 23:23:09 UTC.
Drop events found2,2451,583 (1,360 after de-duplication)Not run: no event-driven backtests
Average return, buy at the close+13.31%+10.1%Not run
Average return, wait three days+12.84%+11.8%Not run
Median return, buy at the close+7.64%+5.0%Not run
Median return, wait three days+9.19%+5.1%Not run
Limits
The answer depends on the drop threshold and the holding period, so one setting does not settle the general question. Stocks that were later delisted are missing from some datasets, which flatters buying dips. A backtest describes the past. The two platforms disagree on which entry had the higher average: NexusTrade found 2,245 events and favored buying at the close, QuantConnect found 1,583 and favored waiting. They drew S&P 500 membership from different sources and QuantConnect's prices include dividends.

The results

Observed

NexusTrade: what it returned

Return over the next 20 trading days, 2,245 drops of 10% or more

EntryAverage returnMedian returnShare with a gain
Buy at the close on the day of the drop+13.31%+7.64%68.29%
Wait three trading days, then buy+12.84%+9.19%72.83%

NexusTrade found 2,245 events from January 1, 2020 through December 31, 2025. Buying at the close had the higher average; waiting three days had the higher median and a gain more often.

NexusTrade's chat showing the question, its written answer and a chart of the two average returns
NexusTrade's answer as it appeared in the chat.

QuantConnect: what it returned

Return over the next 20 trading days, 1,360 drops after removing overlaps

EntryAverage returnMedian return
Buy at the close on the day of the drop+10.1%+5.0%
Wait three trading days, then buy+11.8%+5.1%

QuantConnect found 1,583 events and kept 1,360 after removing overlapping ones. Its assistant reported that waiting won by 1.71 percentage points on average, with medians of 4.99% and 5.12%.

From QuantConnect's own reasoning

"Mean 20-day returns are huge (15.7%) - that's heavily influenced by 2020 COVID mega-bounces. Medians are more representative: ~6% for both."

"Deduped: mean A 10.06% vs B 11.77% - waiting 3 days wins by 1.71pp on average with paired t-stat 3.34. But caution: this is driven by tail outliers. Medians nearly identical (4.99 vs 5.12)."

Quoted from the assistant's working notes, lightly trimmed.

QuantConnect's assistant reporting the event study results
QuantConnect's answer to the event question.

Composer: what it returned

Composer said it does not do event-driven backtests

"Composer does not support event-driven backtesting or direct comparisons of specific entry and exit strategies based on price movements. It also lacks the ability to calculate returns based on specific historical events like a 10% drop in one day."

It offered a different strategy in its place: hold SPY when its 14-day relative strength index is below 30, otherwise hold Treasury bills.

Quoted from Composer's assistant.

Composer's assistant answering the wait-three-days question
Composer's answer to the event question.

Comparisons that use this test