Probability Engine
This page is the entry-point for a brief overview about the Probability Engine, its components and architecture, and how they serve the protocol's mission.
Introduction
To underwrite, price, and support credit offerings, Concrete analyzes the inherent topical collateral risk, the initial loan structure, and subsequent credit policy terms, and balances these components against the aggregate internal credit obligation and risk held by the Concrete protocol.
The mission of the Probability Engine is to score the above components as accurately as possible using Data Science and Machine Learning technologies; and empower the protocol to perform safely and profitably.
Our methodology combines classic statistical and quantitative techniques with modern ML, Neural Nets and latest innovations in AI; with a constant focus on conservative back-testing, simulation and monitoring; to ensure all models are measuring risk accurately and robustly, even against the worst black swan events.
System Architecture
Broad architecture diagram
Main components
The Probabillity Engine is composed of the following modules:
Collateral Scoring: predicting volatility or future returns for a collateral token T, on a time horizon of N days, with an emphasis on the downside price action.
Loan & Policy Scoring: evaluating an individual loan and (potential) policy with respect to the collateral type, the internal protocol risk and the particular features of the originating wallet.
Internal/Portfolio Metrics: monitors protocol liquidity and outstanding claim levels.
Decisioning Strategy: the hub that will receive policy applications, run all models and risk assessments; then make approval decisions.
Pricing Strategy: responsible for proposing policy terms and pricing.
Theory
The whitepaper describes the approach and the components above in detail from the theoretical perspective.
Implementation
For implementation, the codebase is composed of the following repositories so far:
ds_data_etl: Data Pipelines from external data sources into BigQuery and Postgres databases.
ds_collateral_features: Feature engineering code to produce the factors used by collateral models.
ds_collateral_models: Models used to analyze and predict collateral forecasts.
ds_airflow: Orchestration of the different components and automation of periodic pipeline runs.
More on tech stack and frameworks here
Integration within the Product Ecosystem
The Probability Engine services will process requests from the backend and publish scores, policy decisions and pricing quotes using a Pub/Sub framework.
This integration is not implemented yet.
Last updated