How Many Samples Do I Need?

When you’re doing a study, one of the first questions you’ll have is “how many samples do I need to get statistically significant results?”, meaning how many trading events do you need to collect such that the statistics from them can be relied upon going forward.

When I had physics lab at Rutgers our instructor took us through the math to show we needed to repeat an experiment for a number of times greater than 30 and less than 36. He gave us an exact number but I don’t remember what it was but I think it was 32 or 34. I do remember thinking to myself I’ll use 36 because that’s my favorite number.

I wasn’t sure if you could apply physics experiment statistics to financial market behavior. I tried to find the answer online, but after hours of searching I didn’t find it.

I called into Larry Pesavento’s internet show on tfnn.com and asked Larry’s guest Stan Harley (www.harleymarketletter.com). He was an aeronautical engineer and currently offers analysis and cycle research in the financial markets. He said the number is about 30, which confirms my memory from physics lab.

Supporting a minimum sample size of 30, here is a quote from https://www.investopedia.com/terms/c/central_limit_theorem.asp#citation-6

Why Is the Central Limit Theorem's Minimize Sample Size 30?

A sample size of 30 is fairly common across statistics. A sample size of 30 often increases the confidence interval of your population data set enough to warrant assertions against your findings.

The higher your sample size, the more likely the sample will be representative of your population set.

However, I believe more is better with regard to sample size, but only up to a point. At some point you’ve reached the limit of diminishing returns.

But you need to consider cycles in the market. There are all kinds of market cycles. The markets go through different modes of behavior, which can affect trading strategies. So, to get statistics that you can rely on over time, you’d want to collect samples through every cycle possible. This could affect the number of samples.