Quantitative Portfolio Management

Chapter 3Forecast Combining

1. Alice and Bob own bad watches telling time with independent random errors. Alice's watch says it is 1 pm, and Bob's 2 pm. What time is it?
2. Alice and Bob are quant analysts at a hedge fund. They independently develop forecasts from unrelated data sources. Asked about the AAPL stock return for tomorrow, Alice's forecast is 1 bps, and Bob's 2 bps. What is the forecast for AAPL?

From a quant interview

Given the multiplicity of financial datasets and forecasting ideas, quants soon arrive at a library of forecasts to be used for trading. A “clean” way of trading a separate book for each forecast is more common than one might think, but there are clearly better ways. For example, crossing opposite trades from such books internally will save on trading costs, but this is still suboptimal because individual books would forego many trading opportunities seen as not worth the cost.

Combining weak forecasts into a potentially stronger forecast (when pointing in the same direction) will make more profitable trades possible. In the case of netting (opposite sign) forecasts there is saving in terms of cost of capital, which is often expressed as a limit on the portfolio gross market value (GMV).

Combining alphas for portfolio trading is another kind of alpha. The question of how to combine forecasts has been raised in other fields as well, most frequently in econometrics.¹ Econometric forecast combining almost always means equal-weight or a similar static averaging ...