tecton

Personal project to build a quant investment platform while testing out some newer tools/libraries.

• dagster for data ingesting, pipelines, job scheduling
• ibis with duckdb backend for data querying
  • store data in parquet

Abstractions

• Mantle: core data access
• Markets: a collection of tradeable futures markets
  • Market: entries of Markets
    • some metadata is lazy-loaded
  • Contract: the discrete futures contract
• FactorModel: factor model instance (SignalModel)
  • ModelDefinition:
    • contains necessary information to build the model
  • types:
    • AlphaModel
    • RiskModel
• Optimizer:
  • set up a generic problem and solve it
• Strategy:
  • given a model, initial holdings, get optimal portfolio + trades
• Simulation:
  • backtest of a Strategy.
• Portfolio:
  • collection/time series of holdings.
  • perform various analytics via standard interface.

Vision for Supported Asset Classes/Strategies

• Futures+Forwards
  • Trend
    • price trend
    • fundamental trend
  • Carry
• Crypto
  • Fundamental L/S
• Equities
  • Fundamental L/S

Notes on env setup:

Install uv

curl -LsSf https://astral.sh/uv/install.sh | sh
brew install ta-lib
source ~/.local/bin/env
uv sync

.bashrc

source .venv/bin/activate

ta-lib

https://github.com/TA-Lib/ta-lib-python

Target Data Calls

from tecton.dal.mantle import Mantle

Mantle.select(Mantle.Tables.futures.cont, start, end, columns=[])

TODOs

laptop setup

• copy raw data files
• implement S3 and local storage options

Futures Data

• add data checks for partitions
• return data sorted...

Signals

• store model definition in yaml
• build factor-forecast signal pipeline

Portfolio Construction

• calc/store vol
• try using garch model for vol
• calculate market and asset correlations

Features

• lookback capability

Design Decisions

• yamls live close to code (instead of having a central "config" dir)