1
$\begingroup$

I am sure there are a lot of papers that are related to stock market analysis.. but I haven't been able to find ones that fit my needs most. I want to read papers, replicate their analysis, and use them for my current job. Specifically, I am seeking to build a model that

1) given a timeframe, determines which factors (sectors, regions, styles, etc) influence the stock market, say S&P 500 or MSCI ACWI. The number of factors can be as many as 100.

2) uses some machine learning techniques.. other than simply linear regression..

3) not too difficult to implement.

Thank you for your help in advance!

Improve this question
$\endgroup$
2
$\begingroup$

What you are looking for is a decomposition of stock/index returns. You can do this via a PCA approach or Beta approach.

A well-known PCA model is Sungard's APT model (to provide just one link: APT Modelling Guide). You can find plenty more resources by googling.

Also have a look at this Quant SE post: How to use PCA for trading.

Improve this answer
$\endgroup$
5
  • $\begingroup$ Thank you so much. I think Sungard's APT model seems very interesting. I will go ahead read it soon! (the pdf file is very long though haha.. 350 pages...) $\endgroup$ – Jun Jang Jul 10 '18 at 11:31
  • $\begingroup$ for my goal, I guess I don't need to read the whole thing, right? lol $\endgroup$ – Jun Jang Jul 10 '18 at 11:38
  • $\begingroup$ Sungard's 50-Factor Model is just one example of PCA. If you haven't done PCA before, I recommend you read the wiki article first or look at coursera, e.g. coursera.org/lecture/fundamentals-machine-learning-in-finance/…, and then write your own first implementation (maybe start with a 2-factor PCA model). $\endgroup$ – Phil-ZXX Jul 10 '18 at 11:49
  • $\begingroup$ I have done PCA to compute Mark Kritzman's absorption ratio, a measure of systemic risk. I think one problem with PCA is that the linear combinations of the asset returns may be hard to interpret, in terms of economic/financial meanings. $\endgroup$ – Jun Jang Jul 10 '18 at 12:02
  • $\begingroup$ You can also self-define some of the factors (= force the first e.g. 5 factors to be some combination of major indices or fx pairs like SPX, EURUSD and then let the rest be pure PCA). That way you retain some meaning. $\endgroup$ – Phil-ZXX Jul 10 '18 at 12:12
-1
$\begingroup$

I suggest you read the interesitng paper "Empirical Asset Pricing via Machine Learning".enter link description here

Improve this answer
$\endgroup$
1
  • $\begingroup$ Hi, thank you for the suggestion. I am not really interested in predicting the future returns, which I believe the paper is about. I am more interested in explaining the broad stock market (MSCI ACWI or S&P 500) return with factors. $\endgroup$ – Jun Jang Jul 9 '18 at 11:33

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.