Atium

Quantitative portfolio construction and backtesting framework built on Polars and CVXPY.

Atium provides a modular pipeline for alpha-driven portfolio optimization: define data providers, build a risk model, optimize weights, apply trading constraints, and backtest with transaction costs.

Installation

pip install atium

Quick Start

Single-Date Portfolio Optimization

Construct an optimal portfolio for a single date using mean-variance optimization with trading constraints.

from atium.risk_model import FactorRiskModel
from atium.optimizer import MVO
from atium.objectives import MaxUtilityWithTargetActiveRisk
from atium.optimizer_constraints import LongOnly, FullyInvested
from atium.trade_generator import TradeGenerator
from atium.trading_constraints import MaxPositionCount, MinPositionSize
import datetime as dt

date_ = dt.date(2026, 2, 13)

# Build a factor risk model
risk_model = FactorRiskModel(
    factor_loadings=factor_loadings_provider.get(date_),
    factor_covariances=factor_covariances_provider.get(date_),
    idio_vol=idio_vol_provider.get(date_)
)

# Optimize portfolio weights
optimizer = MVO(
    objective=MaxUtilityWithTargetActiveRisk(target_active_risk=.05),
    constraints=[LongOnly(), FullyInvested()]
)
weights = optimizer.optimize(
    date_=date_,
    alphas=alphas_provider.get(date_),
    benchmark_weights=benchmark_provider.get(date_),
    risk_model=risk_model,
)

# Apply trading constraints
trade_generator = TradeGenerator(
    constraints=[MinPositionSize(dollars=1), MaxPositionCount(max_positions=10)]
)
constrained_weights = trade_generator.apply(weights=weights, capital=100_000)

Full Backtest

Run a backtest with weekly rebalancing, transaction costs, and trading constraints.

from atium.risk_model import FactorRiskModelConstructor
from atium.optimizer import MVO
from atium.objectives import MaxUtilityWithTargetActiveRisk
from atium.optimizer_constraints import LongOnly, FullyInvested
from atium.trade_generator import TradeGenerator
from atium.trading_constraints import MaxPositionCount, MinPositionSize
from atium.backtester import Backtester
from atium.strategy import OptimizationStrategy
from atium.costs import LinearCost
import datetime as dt

start = dt.date(2026, 1, 2)
end = dt.date(2026, 2, 13)

# Build risk model constructor (provides a risk model for each rebalance date)
risk_model_constructor = FactorRiskModelConstructor(
    factor_loadings=factor_loadings_provider,
    factor_covariances=factor_covariances_provider,
    idio_vol=idio_vol_provider
)

# Define strategy
strategy = OptimizationStrategy(
    alpha_provider=alphas_provider,
    benchmark_weights_provider=benchmark_provider,
    risk_model_constructor=risk_model_constructor,
    optimizer=MVO(
        objective=MaxUtilityWithTargetActiveRisk(target_active_risk=.05),
        constraints=[LongOnly(), FullyInvested()]
    ),
)

# Run backtest
backtester = Backtester()
results = backtester.run(
    start=start,
    end=end,
    rebalance_frequency='weekly',
    initial_capital=100_000,
    calendar=calendar_provider,
    returns=returns_provider,
    benchmark=benchmark_provider,
    strategy=strategy,
    cost_model=LinearCost(bps=5),
    trade_generator=TradeGenerator(
        constraints=[MinPositionSize(dollars=1), MaxPositionCount(max_positions=10)]
    )
)

print(results.summary())
results.plot_equity_curve('equity_curve.png')

Modules

Module	Description
Data Providers	Calendar, Returns, Alphas, Factor Loadings, Factor Covariances, Idio-Vol, Benchmark Weights
Risk Model	Factor risk model estimation and construction
Optimizer	Mean-variance optimization (MVO) with pluggable objectives
Objectives	MaxUtilityWithTargetActiveRisk and others
Optimizer Constraints	LongOnly, FullyInvested, etc.
Trade Generator	Post-optimization trading constraints (MaxPositionCount, MinPositionSize)
Strategy	OptimizationStrategy, QuantileStrategy
Cost Model	NoCost, LinearCost
Backtester	Time-series backtesting with configurable rebalance frequency

To Do

• Signal Combinator
  • Equal
  • Inverse Volatility
  • Fama-MacBeth
  • Elastic Net