Recency bias - It is not solved by being quant

The recency bias states that we more easily remember or recall events which have occurred more recently than those in the past. Thinking about this bias leads to a deeper discussion on how far back in time should your memory go. Or, how much memory do you need to avoid this recency bias? 

This is especially important when considering investment decisions. How far back in time should you look for answers of diversification and return prediction? What are recent and what are distant memories and does recency bias change based on the problem to be addressed? 

The power of any recency bias is not avoided just because you use quantitative models. In fact, recency issues may be worse with data driven strategies because decisions are solely based on sample sizes that may be limited. In the general case of using use mean and standard deviation of return, all sample data is equal but there is a bias based on the sample chosen. By definition, using a given sample states that you have no memory after the initial data sample used. The look-back period is critical with accepting or rejecting a specific strategy. For example, a look-back of five years has a recency bias for the last five years. Events beyond five years have no relevance for the analysis.

More data is generally better than less data; nevertheless, inference using daily data for five years may be acceptable for some decisions. However, if we are making a strategic allocation decision, there is a recency bias even if you look over just a single business cycle. In the asset allocation decision, there may be a need for decades of data if it is available. At the other extreme, there can be data saturation if too much information is used, and the economic environment has changed. 

Any recency bias is decision relative. An "unbiased" sample for one decision is very biased for another. Being quantitative does not change the fact we can be data biased. Appreciating a flexible definition of recency bias is useful.