Concept

Alpha

One way to conceptualise an alpha is to outline its processes and model our understanding around its behaviour. The figure below depicts an overview of how an alpha behaves particularly on how it process data and eventually turn them into trading signals.

$x$ = input data
$y$ = model data

Model and Signal Generation in this scenario can be thought as mathematical functions albeit their signatures are not necessarily the same. Note that $x$ can be any data (for eg. market data, on-chain events, etc.) and the exact structure of $y$ is not restricted, it is up to the implementor to decide how to translate $y$ into signals later.

Model

Model can be any of the following functions:

• $f(x) \to y$
• $f(x) \to y_1, \;y_2, \;y_n$
• $f(x_1, \;x_2, ... \;x_n) \to y$
• $f(x_1, \;x_2, ... \;x_n) \to y_1, \;y_2, ... \;y_n$

Note that from the possible signatures, it implies that $x$ can be a singular data or multiple data hence it is possible for an alpha to utilise multiple data sources as input to its model. For example, a simple model

$y = x_1 - x_2$

$x_1$ = BTC price on Exchange A
$x_2$ = BTC price on Exchange B

In this case, the model computes the difference in price across Exchange A and Exchange B.

Signal Generation

Signal Generation can be any of the following functions:

• $f(y) \to [-1,1]$
• $f(y_1, \;y_2, ... \;y_n) \to [-1,1]$

A trading signal to put simply is a real number ranging between $-1$ to $1$ where negative number implies a sell position, positive number implies a buy position and $0$ is a special number to indicate empty position.

The input of the signal generator fits specifically to how one's $y$ (model data) is structured hence why it can take $y$ or $y_1, y_2, ... y_n$. The only specification is that it must return a number between $[-1,1]$ to indicate its confidence level on future price going up / down.