I'm building a pattern recognition model for my master thesis. The idea is to build a framework with some Macro variables (long/short term rates; rates differential; equity; fx; vix) in order to find wich asset class (or investment style or strategy) would perform better on the current period, based on similarities with historical data. For that purpose I am using the K-nearest neighbour algorithm. I would like to ask sugestions regarding not only the quantitative method (KNN) but also the most significant macro variables to use.

I also would like to ask if you know any relevant literature regarding this or any similar theme?

Thanks in advance

  • $\begingroup$ Why do you call it analog? Anyhow I don't think kNN is a good tool for this purpose as the variables will have different distributions the distance metric you will use will have very unintended results. $\endgroup$ – Cowboy Trader Sep 18 '14 at 9:33
  • $\begingroup$ Hi! Thank you for your answer. I call it analog since I will be trying to compare patterns across different regions. In order to avoid the problems you've mentioned I have standardized the data in order to avoid unintended results $\endgroup$ – goncalogc Sep 18 '14 at 9:36
  • $\begingroup$ Standardization will help if they are from the same type of distribution but with different parameters. If the distributions are of different type standardization will be ad-hoc. Also most of the time there will be variables which will have to be excluded. But kNN includes all of the variables. To test this you can simply generate a bunch of random numbers and introduce these as new variables and you will see that they degrade performance. How do you know that your candidate variables are all useful? $\endgroup$ – Cowboy Trader Sep 21 '14 at 17:01

Such an approach is done by the systemic investor blogger in his blog Time Series Matching with Dynamic Time Warping.

  • $\begingroup$ Thank you very much Richard! I'll take a look on that for sure $\endgroup$ – goncalogc Sep 18 '14 at 13:52

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.