2
$\begingroup$

The above ensemble methods appear useful in several machine learning competitions, like Netflix prize or KDD. They work by diversifying between several model variants.

Are they also useful in portfolio construction or other quant finance subjects? What are examples of their uses?

$\endgroup$
  • 1
    $\begingroup$ Not an example of these specific methods, but you can think of the general approach as accounting for parameter and model uncertainty (i.e. approximately integrating out the model and parameters from the posterior predictive distribution). It is not typically motivated in this way but stochastic volatility models (for option pricing, say) accomplish this goal by accounting for uncertainty in the volatility parameter. $\endgroup$ – Chris Haug Dec 17 '16 at 18:12
  • $\begingroup$ Interesting to think about these methods in the Bayesian framework. What surprises me still is that there is a huge literature on Bayesian methods in finance and very few on the ensemble approach. The Bayesian method can have analytical and computational benefits (which definitely helps in risk neutral pricing), but would it really forecast that much better in the physical measure? $\endgroup$ – Mark Horvath Dec 17 '16 at 21:28
  • 1
    $\begingroup$ The idea doesn't depend on the measure. You should account for parameter and model uncertainty to compute VaR correctly ("model risk"), for example. Another one is accounting for regime-switching or unexpected structural breaks after the end of the sample. Pesaran et al have a paper where they show better forecasting performance on US T-bill rates by including uncertainty about the number of future breaks, and the parameters in each of those regimes. $\endgroup$ – Chris Haug Dec 17 '16 at 22:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.