Apr

28

This principle outlines the idea that markets change their behaviour. For example:

- the 13th business day of the month made money for 5 years ( lost for some days in the year, but overall ended the year profitable)

then the market changes and

- the same 13th business day of the month loses money for 5 years ( won for some days in the year, but overall ended the year at a loss)

If you tested the data set for the last 10 years as a single block, (5 years losing + 5 years winning) the regularity would completely cancel out. The only thing I can think of is testing data sets in chunks of say 1 year, but over 10 years. and avoiding large blocks in general. Any suggestions?

Steve Ellison writes:

I agree with the idea of shortening the lookback period. I remember the Chair suggesting 60 days as a typical lookback period, but obviously you would need a much longer lookback for an event that only occurred once per month.

I would also suggest assessing, if there really is an edge here, why? In this example, which market participants have reasons to sell (buy) on the 11th and 12th days of the month that override price sensitivity, such that others come in to buy (sell) on the 13th day? There is a lot more noise than signal in market price movements, and some of the noise may appear statistically significant if you torture the data long enough, but without an underlying reason, a regularity is unlikely to continue.

Jeffrey Hirsch responds:

The 11th and 12th trading days of the month – what we refer to in the Almanac as the mid-Month spike is driven by automatic payroll deductions bi-monthly or every 2 weeks into 401Ks and other retirement accounts where fund managers are required by law to invest. I discussed this with Steve Eisman on his podcast, The Real Eisman Playbook, recently where he comes in and helps me prove the point. It starts at the 19-minute mark and runs for about 2 minutes.

20:37: mutual funds are only allowed to have a certain amount of cash. So generally, let's say 5%. They're not allowed to have more than 5%. So, if all of a sudden people are putting money in because of their 401k, they got to buy. It's not optional.

Asindu Drileba adds:

I was playing with Google's decision tree Python library called YDF. (it's shockingly easy to use)

I simply used days of the month as features against, up or down days. Mid month dates showed very strong regularities (some up to 60% up days). I never understood why they occurred, until I remembered Mr. Hirsch talking about payrolls hitting brokerage accounts. What's different is that I found these in crypto. But it was harder to trade than I thought.

Larry R Williams writes:

It's been that way for years: the eleventh trading day of the month in stocks is usually a rally day, going way back to the 1980s when I first started trading this.


Comments

Name

Email

Website

Speak your mind

Archives

Resources & Links

Search