Classical Asset Allocation — Offensive (CAA-OFF)
Developed by Keller, Butler & Kipnis · Markowitz + Momentum · Med Risk
Classical Asset Allocation was published by Wouter Keller, Adam Butler, and Ilya Kipnis in their 2015 SSRN paper (#2606884), bridging the gap between tactical momentum strategies and classical Markowitz mean-variance optimization. Where most tactical strategies use momentum to select assets and then apply simple weighting schemes (equal weight, score-proportional, or binary), CAA feeds momentum-derived expected returns directly into a Critical Line Algorithm (CLA) optimizer to produce weights that explicitly target a desired volatility level. The offensive variant targets higher portfolio volatility, accepting larger drawdowns in pursuit of higher returns.
The Critical Line Algorithm, originally developed by Markowitz as a computationally efficient method for solving mean-variance portfolio optimization problems, requires two key inputs: expected returns for each asset and a covariance matrix describing their joint behavior. CAA's innovation is using trailing multi-period momentum composites as the expected return estimates — leveraging the empirically documented persistence of momentum to generate forward-looking return expectations from backward-looking data. This approach grounds the classical optimization framework in the same momentum anomaly that drives simpler tactical strategies, but adds the dimensional benefit of considering correlations and volatilities in the allocation decision.
The strategy operates on an eight-asset universe spanning US equities (SPY), international developed (EFA), emerging markets (EEM), US tech (QQQ), real estate (VNQ), gold (GLD), intermediate bonds (IEF), and short-term Treasuries (BIL). Each asset's expected return is estimated as its 13612U composite — the average of one, three, six, and twelve-month trailing returns. The covariance matrix is estimated from trailing twelve-month daily returns. These inputs are fed into the CLA optimizer, which calculates the portfolio on the efficient frontier that matches the offensive target volatility.
The offensive parameterization targets a higher portfolio volatility than the defensive variant, allowing the optimizer more freedom to concentrate in high-momentum, higher-volatility assets. This produces portfolios with greater equity concentration during bull markets and more aggressive positioning overall, at the cost of larger drawdowns during periods when the momentum signal is incorrect or lagging.
How It Works
Momentum-Derived Expected Returns
Each month, all eight assets are scored using the 13612U composite — the unweighted average of their trailing one, three, six, and twelve-month total returns. These scores serve as the expected return inputs for the CLA optimizer. An asset with a 13612U score of 8% is treated as having an expected monthly return proportional to 8%, while an asset scoring -2% is assigned a negative expected return.
The multi-period averaging provides more stable return estimates than single-period returns. An asset that has been consistently strong across all four timeframes receives a higher expected return estimate than one showing strength in only one or two periods, reducing the optimizer's tendency to chase short-term noise. The 13612U composite has been validated across multiple tactical allocation studies as a robust predictor of short-to-medium-term asset class returns.
Critical Line Algorithm Optimization
The CLA optimizer takes the momentum-derived expected returns and a trailing twelve-month daily covariance matrix as inputs and computes the efficient frontier — the set of portfolios that deliver maximum return for each level of volatility. From this frontier, the optimizer selects the portfolio matching the offensive target volatility.
The CLA's advantage over simpler weighting schemes is that it considers the full correlation structure among assets. Two assets with identical expected returns but different correlations with the existing portfolio receive different weights — the one providing more diversification receives higher allocation. This correlation-awareness prevents the portfolio from inadvertently concentrating in highly correlated positions that look diversified by name but behave as a single bet in practice.
Target Volatility and Risk Management
The offensive variant's higher target volatility gives the optimizer more latitude to concentrate in high-momentum assets, producing portfolios that can allocate forty to sixty percent to a single asset class when that class shows both strong momentum and acceptable risk characteristics. During periods of uniformly high momentum across equities, the portfolio may hold predominantly equity positions with minimal bond allocation.
When momentum shifts or market volatility spikes, the optimizer automatically adjusts. Rising volatilities and changing correlations cause the optimizer to reduce exposure to affected assets even before their momentum signals turn negative, providing an implicit early warning system embedded in the optimization process itself. This is a key advantage over simpler momentum strategies: the portfolio adapts to changing risk conditions through the mathematics of optimization rather than relying on threshold-based signals.
Explore Classical Asset Allocation — Offensive (CAA-OFF)
See the full backtest across 18 years of market data, or run your own what-if scenarios by adjusting all research parameters.