Return-based & Holding-based Style Analysis

Reading 33 Return-based Style Analysis can be excuted rapidly & Holding-based Style Analysis can detect style drift quickly ? It seems discrepant ! Can anyone advise ?

returns based is just factor analysis (regression). it’s quick and easy holdings based can detect drift faster than returns based because it uses only recent data. returns based analysis uses mostly past data so it is heavily influenced by history and is slow to show drift.

Returns based analysis is over a period of time (think a movie). Holdings based is at one point in time (think a picture). Since returns based analysis acts like a moving average it will take awhile to show style drift. Holdings based can see it right away when the next picture is taken.

1 Like

Return base analysis, you could run it as a regression on a spreadsheet (which can be executed rapidly) with various styles as the coefficient. Since it’s based on historical return, it doesn’t detect style drift as quickly. Whereas in holding based, you actually see what the portfolio is made of, at the moment of analysis, hence you could catch manager’s style’s drift more quickly.

But why Holding-based Style Analysis needs more data than Return-based Style Analysis ? Return-based Style Analysis uses historical data to regress and needs more data, right ?

Historical return data can be easily obtained from various sources But you’d need to sit down, and subjectively rate the characteristic of each holding in a holding based style analysis.

Returns based only needs the returns of the portfolio in question and the returns of the style indicies. Holdings based needs the data for all the underlying securities on the portfolio.

1 Like

What is the appropriate amount of data needed for returns-based analysis? I think I’ve seen 3 years of at least quarterly data? Does that sound right?

ilvino Wrote: ------------------------------------------------------- > What is the appropriate amount of data needed for > returns-based analysis? I think I’ve seen 3 years > of at least quarterly data? Does that sound > right? usually like 3 year rolling 1 year, or rolling 3 year returns