10: Forecast Model Accuracy and Precision – Part 3

Disclaimer: The Content is for informational purposes only. You should not construe any such information or other material as legal, tax, investment, financial, or other advice. To learn more, read our Terms of Use.

Made a few more fixes to the algorithm:

  • Noticed that the model had the capability to produce negative price forecasts. I’ve removed any instances that this occurs
  • Because of negative values, it also caused the potential where the “low prediction” can end up higher than the average forecast. Fixing the 1st point also fixed this issue

Focusing back to reducing forecast error, my first hypothesis was that volatility was the key reason why the accuracy was so low in some stocks. Unfortunately, reviewing the results, volatility wasn’t the reason.

I started to dissect each stock that have high forecast error and the conclusions are the same. Usually this happens when the stock are impacted by macro events that cause the company to experience significant growth/declines.

For example, AMD went from a 0.75x PB value in 2014 to and 10x PB in 2021 and my forecast is still yet to catch up.

Another example was Tesla, that experienced similar results. But for Tesla, the growth was in a shorter time span. Tesla’s PS ratio went from a 2x in 2019 to 16x in 2021. That’s a 8x increase in a span of 2 years.

The conclusion is this. The forecast works very well for blue chip mature companies that are growing consistently on an annual basis. The forecast however does a bad job in sniffing out underpriced sectors and understanding the macro environment.

There are no variables in the model that tracks the explosion growth in the semi-conductors sector or the demand for electric vehicles. Therefore, in order to improve on the forecasts at the individual stock level, I have to segment my analysis at the subsector level and apply multiples based on the macro economic environment.


Please enter your comment!
Please enter your name here