Why predict?

1/ Many people were following JH during his years on the web and consensus is that he has some good ideas but makes trading too complicated.

2/ There is a matrix as described above but I doubt that JH ever written about it or knows enough to understand it.

 

x2

 

So, it is a case of forecast that requires skill and one that does not i.e. a case of “Luck Vs Reasoning & Statistics”.

Let us take a few simple examples, to clarify ourselves:

a) Weather Forecast:

Taking the weather forecast, of whether it shall rain or not. Now, we know that it rains 33% of the time in New York. So if we say that it will not rain, we shall be 67% right. But, this can not be called a “forecast with skill” as it is based on probability/guess.

We can also do the same by putting 20 pieces (with “No Rain” written on them) and 10 pieces (with “Rain” written on them) in a hat. Then, shake the hat and take out one. This too can not be called a “forecast with skill” as it is based on luck.

But, by doing the same forecast with the help of weather maps and models of the atmosphere, we are using some reasoning and statistical methods. So, this can be considered as a "forecast with skill”. Moreover, this forecast should be better than the above two or else is of no use then.

b)Trading:

Now, if we wish to predict the “Tomorrow High”, we can use the following simple test, which requires no skill and is a simple “Standard Forecast”.

Tomorrow High=Today High + (Today High-Yesterday High)

In order to make a more accurate forecast i.e. “forecast with skill”, we need to apply statistical or reasoning methods to remove any luck/error factors.

We can better the above forecast to the following:

Tomorrow High = (Tomorrow Open) + (0.5*Average (Range, 40))

But, this too may not be a skillful forecast, though a better one when compared to the above standard one.


Applying the "Lower Root Mean Square Error" to test out whether our statistical or reasoning methods are having some “skill” or not and can be considered as “forecasts with skill”

We can also make use of the "Multiple Regression Equation"; a modeling method which may or may not show skill.


The main point is to create a “standard that requires no skill” and then using statistical or reasoning methods, to create “skillful forecasts”. Now, we need to compare the above two and make sure that the “Skill forecasts” better the “No Skill” forecasts.

 

Good post, Skill score concept is very big in weather forecasting and is very useful for market prediction.

In market prediction you need to square the difference, We want to punish large errors more than small ones. We need to do this both for the standard as well as our predictions.