Reversion to the mean is typically used in retrospect to describe an event that has already happened. Using it for a forward-looking calculation is almost always destined for failure, especially in this case.
When data has a simple history of nice clean data around a fairly clear trend line (i.e. if the mean is the same no matter how far back in time you go), using reversion to the mean for a forward-looking estimate makes some sense.
However, if the mean of the last 5 years is different from the mean of the last 10 years is different from the mean of the last 50 years, the results of your reversion to the mean calculation are going to be basically random numbers.
Picking too short a timeframe means you don’t have a large enough number of samples. It also can ignore long term market effects.
Furthermore, the mean can revert to the data as well if your timeframe is too short. If I am looking at annual data and I choose a timeframe of 5 years, my next two datapoints are going to have a significant effect on the mean, even if the next two data points are the same as my 5th data point. If you wait long enough, eventually the mean will get very close to the data, even if the data does not change at all.
Picking too long a time frame ensures you have enough samples, but can underestimate the effect of any unusual short-term, current market conditions.
Secondly, picking the right variable is critical – is it prices? price growth? price growth relative to interest rates? prices relative to income? Interest rates themselves? Some combination of price, income and interest? Not all these variables can simultaneously revert to their means because they play off each other.
In the case of housing in the US, the tax anomaly in 1997 could have a drastic effect on the mean, so you have to be careful looking at data before then. But, that really isn’t enough time to get an understanding of what the mean is under these conditions. So, this is just an awful market in which to use this technique.
Finally, when data is random, we don’t know when the reversion will happen.
So, yes at some random time, something will revert to the mean of the last x years. Alot of good that does us.
Reversion to the mean is an analysis performed when they analyst doesn’t really know very many analytical techniques.
