Comments about the DLA and Probabilities

Given that the DLA seems to be the hot item "du jour" I'm getting many questions about the correct ways to interpret the probabilities displayed on the table. So to clarify misconceptions once and for all I want to post a basic premier about probabilities in general first and then more specifics about the DLA itself.

Let's start with a quick review of the following question: Is the market going to move 1% up in 5 trading sessions ? (That is the basic question that the DLA wants to answer). The answer can be a categorical YES, or a categorical NO.

If we get a YES answer then the meaning is unambiguous is clear that the DLA thinks that the market will move 1% up in 5 trading sessions and we plan accordingly. However what happens when we get a NO ? how can that be interpreted?

If we get a NO answer what can be said of the market ? For starters a NO could mean the following:

1. The market will move up 0.999999999% in 5 trading sessions or less (not quite 1% or more) 

2. The market will remain at the same level.

3. The market will fall.

As you can see a categorical NO answer from the DLA is useless, we can't really say anything about the market in that case (or more exactly what we can say is not very useful or actionable in terms of trading).

With the categorical case out of the way, now let's look at the probabilities. As you know the DLA doesn't spit YES or NO answers, instead it produces a number between 0 and 1 and we make a decision to pick a threshold for YES or NO. Also, the algorithm is not infallible, we know that it can be wrong and from the statistical analysis performed on it for 2017 we can conclude a couple things:

The precision of the algorithm is 64%, what this means is that out of 100 YES signals, 64 of them turn out to be correct. This number is more than enough to produce a massive positive edge when trading the 1% (or more) move up with options. However it is not 100% and you need to be prepared to take losses on that 36% of cases where it is wrong.

Second, the recall of the algorithm is 43%, what this means is that of all of the 1% or more moves up that actually happened, the DLA predicted 43% of them. In net terms it means that the DLA generates a lot of false NO's (57% of the time a NO will be false). Here you can see that the level of false negatives, coupled with the inherent ambiguity of the NO answer makes it worthless to take any trading decision when the DLA signal is "NO". And that is fine because we only trade the "YES" signals.

So as a practical example the DLA is generating a string of "NO"s for next week (very low probabilities) so what can we say about the market? Nothing, the only thing practical is that we are not taking bullish lottos for next week. The market could keep going up a lot (false negative), or remain range-bound or fall. So as you can see the DLA should be used only for taking that bullish lotto trade and nothing else.