How many times have you bought a bottle of milk? And how many times have you bought a house?
Which are you more comfortable with?
It's no surprise that we make better repeated small decisions than the once in a lifetime big choices.
When short horizon strategies make mistakes they dust themselves off and jump back into the market the next day.
With longer horizon strategies making mistakes lead to dire consequences.
(A 'mistake' is when the strategy's weighting logic says to buy before a large crash, or vice versa.)
A day here or there is very important, as stumbling into or out of a Black Swan can be the difference between life and death.
Take the monthly Skewerage strategy for instance.
When I run as of today with 25 years of back testing data I get a Sharpe of 0.6.
However, running it 6 days earlier gives me six days less data, but also half the Sharpe!
What a difference six days make, huh?
"Today's" backtest gets a stronger buy signal before March 09's S&P 500 rally.
The six day's delay in making that decision makes a world of a difference.
One month contributes to a huge chunk of the overall returns and dwarfs every other month.
Many other longer horizon strategies are similarly affected.
For example monthly Dual Momentum (backtested over 55 years) swings between a Sharpe of 0.57 and 0.42.
Essentially we find very different results from the smallest of changes to the initial conditions.
A whiff of chaos, eh?
Not surprising when you are handling infinitely volatile processes (another reason to hedge away extremes perhaps?).
On the one hand I like the ability to conjure new scenarios and parallel worlds. On the other, the new data overlaps substantially with what we have, meaning less statistical applicability.
It's not 100% clear to me how to incorporate this into the Lazy Backtesting IDE.
Also it would require multiple backtests. I.e. 5 backtests for weekly horizons, 21 for monthly and 63 for quarterly etc.
Not too onerous. Nevertheless the system has to be super efficient to process everything in a timely manner.
This test will cull a hell of a lot of strategies which use the traditional 'month-end to month-end' backtesting approach and overfit that data.
Happiness is a warm gun.