22 Days Later

Quants sweat over time more than anything else.

Calendars, holidays and when payments land are scrutinised over and over.

I remember an infamous hedge fund manager making a stink over our inability to capture Middle Eastern weekends starting on Fridays and the trading week starting on Sundays.

Or the guys who pointed out that US close prices occur nearer to Asian opens. Good idea to capture that way in order to avoid a lot of autocorrelation.

Very smart point which never occurred to me before.

Closes are always important, because you can check your figures against other back office systems as a point of reference.

But opens should probably be considered too. If nothing else, the granularity is always handy as you never have enough data.

Of course every so often France would have three or four days off in a row and throw all your nice correlations off for years after.

In any case, I am sketching out a more general Javascript financial framework than the Lazy Backtesting IDE and thought using it to compare weekend returns against weekdays might be fun.

Code is here.

Turns out the standard deviation of Friday close to Monday close returns is 25% higher than other 'contiguous' weekdays (i.e. no holidays in between) for the S&P 500 since 1950.

I am sure others have done this analysis, just cannot recall who.

It may well be justifiable to use 22 (21 + 4 * 0.25) trading days a month as a rule of thumb; it also has a knock on effect for how you calculate volatility (higher Monday-Friday return weight).

(Maybe only for the anal, but that's a prerequisite for financial analysts, right?)

Then again, looking at the 13 week t-bills since 1960 they show a 40% increase on usual daily returns over the weekend!

There are many cases where you cannot avoid scaling by time in the fixed income world, and precision is super important; so the case for adjusting up from 21 days a month is stronger.


You're welcome everybody!

More laborious work to get your calculations just right.