Quant C scores

Introduction

The quant C scores are a family of intuitive and easy to understand aggregate scores between 0 (most bearish forecast) and 10 (most bullish forecast) of Concrete seven quant models with static weighting based on models’ expected performance and business relevance, and dynamic weighting based on market regimes.

Description

There are two categories of C scores:

• Scores_all: are a combination of Concrete seven quant models. Each model forecast is weighted statically by its expected performance, and its relevance to Concrete business, all relative to the other models. Additionally, models are weighted dynamically every minute as a function of the current market regime and the model relevance to this particular market regime

• Scores_st: are a combination of Concrete three short-term quant models. Each model forecast is weighted statically by its expected performance and its relevance to Concrete business, all relative to the other models

The C scores are updated hourly and feature minutes bar resolution.

Each C scores is provided by coin and on an aggregate based (median across the universe).

Methodology

Various assumptions are made in the methodology used to build the C scores. The main steps are:

1/ For each model, derive an exponentially weighted based z-score of the forecasts

2/ For each model, scale from 0 to 10 using a min-max scaler

3/ Pull latest market regimes weights

4/ Build scores_all and scores_st for each model, taking into account static and dynamic weighting

5/ Save latest data into the database

Example

To pull the scores data from the database, simply run the command:

SELECT * FROM modelsdb.scores.{concept}

Where concept is one of {scores_all, scores_st}.

Reminder: to connect to the database, follow the steps explained in Quantitative framework.