Hi /sci/,

For those who have some experience in the field, does quant trading work ? By "quant trading", I mean trading based on numerical signals that are fully included in the prices and volumes, not fundamental analysis.

In the pic, the author assumes that there are two securities $P_t$ and $Q_t$ and that

$\log P_t - \beta \log Q_t = X_t$

Where $X_t$ is a "noise" (let's say a brownian). So the idea is that at a time $t$, if you observe that
$\log P_t - \beta \log Q_t < \log P_{t-1} - \beta \log Q_{t-1} - 2$ so in this case
$X_t - X_{t-1} = d X_t < -2$ ( -2 is arbitrary but let's say it's a 2 sigma number ).
Then we bet that at the time $t+1$ this quantity will increase so we will buy $1 / P_t$ of $P$ , sell $\beta / Q_t$ of $Q$. If everything goes in the right direction, the quantity $X_{t+1} - X_t > X_t - X_{t-1}$ and we make a profilt.

But does it work in real life ? I have the feeling that it's only a matter of speed and models do not work at all. I may be wrong but even if you look for good features, simple models, collect a lot of data, tune your estimators, I can't really believe that this kind of model will work.

Discuss.