Arithmax Research
← Back to Research

Regime-Based Portfolio Strategies: A Comparative Analysis of Complexity vs. Simplicity in Quantitative Asset Allocation

Author: Frankline Misango Oyolo
Date: 2025
Institution: Arithmax Research
Portfolio Optimization Machine Learning Regime Detection Asset Allocation

Summary

This paper presents a comprehensive analysis of three regime-based portfolio allocation strategies tested over different time horizons. We compare: (1) an original regime-detection strategy with Bitcoin (V1, 2019-2025), (2) an improved version with dynamic optimization (V2, 2019-2025), and (3) a long-term test without Bitcoin (2000-2025).

Our findings challenge the conventional wisdom that increased complexity leads to superior returns. Through rigorous backtesting and honest assessment of multiple bias sources, we demonstrate that a traditional 60/40 portfolio outperforms sophisticated machine learning approaches after accounting for transaction costs and implementation challenges.

Key Findings

  • Traditional 60/40 portfolio outperforms ML-based regime strategies after costs
  • Identified and quantified 10 sources of bias that inflate backtest performance
  • Tested over 7 years (with Bitcoin) and 25 years (without Bitcoin)
  • Incorporated realistic transaction costs, slippage, and regime transition costs

Key Mathematical Framework

The strategy employs machine learning-based regime detection combined with dynamic portfolio optimization.

Regime Detection Model:

$$P(\text{Regime} = k \mid X_t) = \text{softmax}(W_k \cdot X_t + b_k)$$
$$\text{where } X_t = [\text{VIX}_t, \text{MA\_ratio}_t, \text{Yield\_spread}_t]$$
Portfolio Optimization:

$$\max_w \left( w^T \mu_k - \frac{\lambda}{2} w^T \Sigma_k w \right)$$
$$\text{subject to: } \sum w_i = 1, \quad w_i \geq 0$$
$$\text{where } k = \text{detected regime}$$
Transaction Cost Model:

$$TC_t = \sum_i |w_{i,t} - w_{i,t-1}| \cdot V_t \cdot c_i$$
$$\text{where } c_i = \text{asset-specific cost (bps)}$$

Algorithm

Regime-Based Portfolio Allocation Algorithm
  • Collect market features: VIX, moving average ratios, yield spreads
  • Train ML classifier on historical data to identify market regimes
  • For each time period, predict current regime using trained model
  • Optimize portfolio weights for detected regime using mean-variance optimization
  • Calculate transaction costs for rebalancing from current to target weights
  • Execute rebalancing if expected benefit exceeds transaction costs
  • Track performance metrics and regime transition accuracy

Full Paper

Your browser cannot display the PDF inline.

Download PDF