Do “hot” numbers beat a random pick?

What did we find? Across 1,214 test draws, the most-frequent-numbers rule matched 845 main numbers, below the 891.9 expected by chance. A random set held fixed at 8, 9, 15, 26, 32 and 44 matched 944, but its p-value rose from 0.0263 to 0.7896 after adjusting for all 30 strategy-and-warm-up comparisons. No strategy showed a statistically significant advantage and none were profitable.

TOTO just had its 1,234th draw since the change to the 6/49 format. Let's commemorate this two decade-long milestone by examining whether frequency-based strategies (i.e. "hot" and "cold" number rules) actually works over this long period of time.

What did we test?

We used all 1,234 available TOTO draws from 9 October 2014 to 22 October 2026. In the first set of experiments, the first 20 were designated as warm-up draws. Then, before each of the remaining 1,214 draws, every strategy selected six numbers using only earlier main-number results. This walk-forward design avoids using a draw to predict itself.

We compared six rules: most frequent numbers across all earlier draws, most frequent in the prior 10 or 20 draws, least frequent across all earlier draws, a seeded random pick that changes each draw, and a random six-number set held fixed throughout the test. The fixed set was generated once and stayed at 8, 9, 15, 26, 32 and 44.

What happened?

A “match” means at least one of the six picked numbers appeared among that draw’s six main numbers. The null hypothesis (H₀) is that every draw independently selects six numbers uniformly from 1–49, regardless of past results; under H₀, none of these rules can predict the next draw better than chance. The one-sided test asks whether a rule got unusually many matches. Its p-value answers: if H₀ were true, how often would we see at least this many matches?

For the fixed set, 944 matches gave a raw p-value of 0.0263, but after accounting for all 30 comparisons its adjusted p-value is 0.7896, i.e. not statistically significant. Overall, no strategy won Group 1, 2 or 3 and they were all in the red.

Does the 20-draw warm-up matter?

Actually the "most/least frequent across all prior draws” rule already has a history that grows after every draw: it uses 20 prior draws for its first prediction and 1,233 for its last, so the most common "hot"/"cold" number strategy is covered. However, it is fair to argue that using so much past data might dilute the effect of more recent trends (e.g. maybe at some point the machines changed and there's no point using some subset of older data).

To take care of this scenario, we reran the comparison with warm-ups of 50, 100, 200 and 500 draws as well. A longer warm-up also means fewer test draws, so the raw hit totals are not directly comparable; the expected total and chance-model p-value are shown for each test period.

Again, varying the warm-up size did not lead to any profitable outcomes.

So, are “hot”/"cold" numbers useful?

Nope! If you enjoy choosing numbers by a personal rule, that can be part of the fun. Just don’t mistake alluring patterns in past results for a promise about what's to come in the next draw.

The plain-English takeaway