How do I understand selection bias ? Any examples please..Thanks

Don't know if this relates however I had a case of selection bias when working with time series data that excluded dead securities. Hence the data included only the successful companies and thus had a survival bias..

I have seen both the terms used in the context of hedge fund index performance measures, usually there is a distinction between selection bias and survivorship bias.Survivorship bias arises when hedge funds are deleted from a hedge fund databases. The deletion can be due to either that poor performing or bankrupt funds stop reporting, or that the fund have attained a desired size and no longer wishes to be open for investment. The deletion causes an upward bias that is partially offset by the, often good performing, funds closing for investment.Selection, or performance, bias arises when a fund chooses to not be included in a database. Motivations for this can be that the fund, during the initiation period, has attained a desired size or that the fund is performing bad and therefore wishes not to be included. These different reasons compensate each other, limiting the performance bias.One could image similar situations in other types of indices.

Selection Bias is very common. If for example I said that the top twenty stocks (by market cap.) in the current FTSE 100 had outperformed over the last 10 years (vs FTSE for example) that might very well be true but the inference is that they may do so in the future is misleading due to selection bias. If you had taken the top twenty, ten years ago some will have done very badly and dissappeared or at least are not in the top twenty now. It is very common for people to do backtests with existing companies (or other assets) and to neglect to include those that have dropped out of the universe in the mean time. This is actually a special case of Selection Bias known as Survivor Bias (I think). Selection Bias can come from other causes too.An interesting case is the now 'common knowledge' that over long (30y) periods that equity markets have outperformed riskfree rates by 5-8% pa over the last 200 years. However most papers showing this neglect to include the markets that dissappeared (Argentina, Russia etc.) which were every bit as blue chip as America or France in their time. To be fair the risk free rate (sovereigns debt?) in those domains did not fair well either so the bias is not completely cut and dried.

far too much of the literature just gives examples of the markets that disappeared without quantifying their relative importancehere are some relative GDP numbers from Maddison for 1913: US 517, Russia 232, UK 225 and Argentina 29Rajan & Zingales figures for stock market cap / GDP for 1913: US 39%, Russia 18%, UK 109% and Argentina 17%hence relative stock market caps: UK 245, US 202, Russia 42 and Argentine 5 - so Russia and Argentina account for less than 20% of the stock market cap of UKtogether with elroy dimson and paul marsh (Triumph of the Optimists) we've collected equity and other returns over the last century for 16 countries (not including Russia and Argentina) and our best guess is that our sample accounts for over 90% of total world market cap

I agree there are few relevant papers that effectively quantitfy survivor bias or selection bias.Do you know any good links to address these issue ? http://www2.standardandpoors.com/NASApp ... &xcd=500is interesting for equities, wich are supposed to be very long term assets. We can track changes in historical composition of s&p since 2000 ! Hence if we know what companies became after they disapear from index composition it will be possible to quantify survivorship bias for US equities.Of course a list of changes since 1950 would be very more usefull. Feel free to post it.