- EdisonCruise
**Posts:**98**Joined:**

I want to calibrate a mulitvariate linear regression model for forecasting asset return. Now I am doing the following procedure. Subjectively picking up observable factors, like interest rate, fx rate, equity index. Download the data from 2007 to 2016. Select a rolling window, with size 1 year, 2 years or 3 years and the number of lags in regression. Starting from 2010, the linear model is calibrated with data 1 year, 2 years or 3 years back then do the out of sample forecasting of asset return one step ahead. If the forecasted return is positive, then buy, else sell. Then record the actually return. By doing this I pick up size of rolling window and the number of lags, which gives max annual return.With this method, I think there are some problems:(1) Overfitting of parameters. If the model just coincidently successfully predicts certain trend of the asset in some period, then the annual return is high. But this cannot be repeated in the future.(2) In Andrew Ng?s machine learning video, he provides suggestions on how to evaluate the model by looking at the training errors and testing errors. But his method seems not fit in this scenario, because the linear regression model keeps recalibrating in each step. So I down?t know how to determine which features, interest rate, fx rate, equity index can really help to improve forecasting, or just adding noise.

You *have* heard of the efficient markets idea, I presume? One version of it might be: trying to predict returns with kitchen sink regressions is futile.

- EdisonCruise
**Posts:**98**Joined:**

May be that's right, but as I know, factor model is still popular.

There are approximately 30,000 RIA firms in the US alone, all supposedlytrying to do something similar. So, your first thought should be: what unique predictivefactors do I know about, or can at least test, that nobody else has tried? If the answer is 'none', why should your regression have any useful predictive power? A second suggestion.Maybe you should cultivate your techniques *as if* you had inside information.In other words, just for testing purposes, assume you had advance knowledge of a company'snext earnings report. (Not just the earnings, but the whole report).Now develop your models. It might be harder than you think. If you get stuck, you might want to sit in an econometrics class for a couple of semesters if you haven't yet done so. My two cents.

Last edited by Alan on April 13th, 2016, 10:00 pm, edited 1 time in total.

- EdisonCruise
**Posts:**98**Joined:**

Thank you Alan. What I can do is only suggestion one.Developing my model based on recent literatures and hope it works.

QuoteOriginally posted by: AlanThere are approximately 30,000 RIA firms in the US alone, all supposedlytrying to do something similar. So, your first thought should be: what unique predictivefactors do I know about, or can at least test, that nobody else has tried? If the answer is 'none', why should your regression have any useful predictive power? In the end, it all comes down to the estimation and the forecasts of the covariance matrix for optimal portfolio allocations. One can have an edge here even if using the same factors...

GZIP: On