(866)-482-2177

sales@alphatheory.com

REQUEST A DEMO

SYSTEM REQUIREMENTS


Please note the following System Requirements. Further, please limit the number of open applications (particularly price streaming applications) while logged in to Alpha Theory™.


Recommended System Specifications
Processor: Dual Core or Quad-Core 2.4GHz or faster
RAM: 4GB+
Browser: Google Chrome 30+
Screen Resolution: 1280 x 1024 or greater
Internet Access: Business Class High-Speed


Minimum System Requirements
Processor: Intel Pentium-M 2.0Ghz or equivalent
RAM: 2GB+
Browser: Google Chrome, Mozilla Firefox, Internet Explorer 9+ (without Compatibility View), Safari
Screen Resolution: 1024 x 768 or greater
Internet Access: High-Speed

Subscribe to Alpha Theory content

Alpha Theory Blog - News and Insights

« October 2018 | Main

1 posts from November 2018

November 02, 2018

Better Predictions Lead to Better Returns (Garbage In – Garbage Out)

 

Not every position is better off following the model position size (optimal) determined by Alpha Theory. However, the times when optimal outperforms are associated with higher forecast accuracy. If you put better forecasts into the model, the model does better. This is a straightforward demonstration of Garbage In-Garbage Out.

 

Correlation of Actual and Forecasted Returns for Positions that Under/Overperformed Optimal

 

Screen Shot 2018-11-02 at 12.09.17 PM 

Models are data dependent. When good data is input in the model, the model has higher predictive power. Bad data in and, well, it doesn’t have the same edge. The correlations hold if we expand into quartiles.

 

Correlation of Actual and Forecasted Returns for Positions that Under/Overperformed Optimal

 

Screen Shot 2018-11-02 at 12.10.57 PM

Picture0 

And largely holds for deciles:

 

Correlation of Actual and Forecasted Returns for Positions that Under/Overperformed Optimal

 

Screen Shot 2018-11-02 at 4.11.15 PM 

Picture1

 

What you’ll notice is that the correlation overall between actual and forecasted returns is fairly small with the highest decile showing an 18% correlation. Even though the signal is faint, it is strong enough to power a model that produces positive returns.

 

As the data shows, it is worth taking the time to measure your historical forecasting skill. If you have positive forecasting skill, then a simple model can dramatically improve results.