Forecasting Tangency Portfolios and Investing in the Minimum Euclidean Distance Portfolio to Maximize Out-of-Sample Sharpe Ratios

Abstract: We propose a novel model to achieve superior out-of-sample Sharpe ratios. While most research in asset allocation focuses on estimating the return vector and covariance matrix, the first component of our novel model instead forecasts the future tangency portfolio, and the second component then determines the optimal investment portfolio. First, to forecast the tangency portfolio, we forecast the efficient frontier by decomposing its functional form, a square root second-order polynomial, into three interpretable coefficients, which can then be used to calculate a forecasted tangency portfolio. These coefficients can be forecasted using vector autoregressions. Second, the model invests in the portfolio on the efficient frontier that is the minimum Euclidean distance from this forecasted tangency portfolio. A motivation for our approach is to address the limitation that the tangency portfolio only maximizes the Sharpe ratio when future returns and covariances are stationary, and can be directly estimated with historical data, which often does not hold in out-of-sample data. Our approach addresses this shortcoming in a novel way by forecasting the tangency portfolio, rather than estimating return and covariance. For empirical testing, we employ two sets of assets that span the market to demonstrate and validate the performance of this novel method.

• The paper introduces a forecasting approach that directly parameterizes the efficient frontier, bypassing traditional return-covariance estimation methods.
• It employs an online VARX model to predict three interpretable coefficients (TMVP, OMVP, and u) that streamline robust portfolio construction.
• Empirical results show that the minimum Euclidean distance portfolio consistently outperforms benchmarks with higher Sharpe ratios and lower drawdowns.

Model-Based Forecasting of Tangency Portfolios and Minimum Distance Investing

Context and Motivation

The instability and estimation errors associated with classical mean-variance optimization (MVO), particularly under conditions where asset returns and covariances are nonstationary, have motivated substantial research to improve portfolio selection methods. Canonical models such as Black-Litterman, covariance matrix shrinkage, and DCC-MV-GARCH provide enhancements over MVO but retain limitations in return vector estimation, model complexity, and practical interpretability. This paper presents a methodology grounded in direct forecasting of the efficient frontier's parameterization—eschewing the standard return-covariance estimation paradigm—and introduces the concept of investing in the minimum Euclidean distance portfolio relative to the forecasted tangency portfolio to enhance out-of-sample Sharpe ratios (2604.03948).

Methodological Framework

Parameterization of the Efficient Frontier

This approach leverages a dimensionality reduction of the efficient frontier, using the established result that it can be represented as a square root second-order polynomial with three interpretable coefficients—TMVP, OMVP, and $u$. These coefficients correspond to the return and standard deviation at the minimum variance point, and a curvature/utility parameter quantifying the usefulness of mean-variance optimization for a given asset set. Their interpretability aids both in model transparency and downstream forecasting.

Time Series Forecasting of Frontier Coefficients

The authors employ an online Vector Autoregression with Exogenous Inputs (VARX), lag order 1, tailored for short-horizon forecasting (21 business days ahead) of the three coefficients. The feature set is intentionally sparse (including prior coefficients and the historical equal-weighted moving average), enforcing a parsimonious model and mitigating overfitting risk while maintaining interpretability. Coefficient regularization is achieved through exclusion of predictors lacking in-sample significance.

Construction of the Minimum Distance Portfolio

Given the forecasted efficient frontier coefficients, the anticipated tangency portfolio (traditionally maximizing the Sharpe ratio under stationarity) is infeasible to directly reconstruct due to unobservable future returns and covariances. Instead, the optimal investment portfolio is selected as that which is the minimum Euclidean distance (in expected return-volatility space) from the forecasted tangency location on the currently estimated efficient frontier. This embeds robustness by adapting to market nonstationarity and inherent estimation uncertainty, offering a closed-form, convex objective amenable to optimization via Newton's method for solution of the distance-minimizing target return.

Empirical Evaluation

The empirical study utilizes two diversified asset universes (Fama-French style growth-value-market capitalization mutual funds plus bonds, and the full range of S&P sector ETFs with bonds) over 33 years. Model training is performed on a rolling out-of-sample basis, with online updates and enforced leverage constraints (max 1.5x). Transaction costs (1% per day on weight changes) are incorporated for realistic performance accounting.

Key empirical results, reflecting time out-of-sample (2000–2022 and 2008–2022 for each universe), include:

• The minimum Euclidean distance portfolio consistently achieves higher Sharpe ratios and lower drawdowns relative to four strong benchmarks: rolling 1-month tangency, equal-weighted, S&P 500 total return, and classic 60/40 stock/bond allocations.
• For GVMC and Bonds: Sharpe ratio of 1.00 vs 0.67 (rolling tangency), 0.41 (equal-weight), and 0.33 (S&P 500). Maximum drawdown is less than half that of the S&P 500 index.
• For Sectors and Bonds: Even under 2x leverage, the model's Sharpe ratio (0.76) surpasses all benchmarks, and it delivers drawdown and return performance superior to both passive and classical active benchmarks.
• Alpha regressions indicate statistically significant excess returns (p-value $<$ 0.01) when compared to all baseline models.

Theoretical and Practical Implications

This methodology obviates explicit assumptions of stationarity in the return-covariance process by recasting portfolio selection as a time series forecasting problem in reduced-dimension parameter space. The explicit interpretability of coefficients facilitates model trust and diagnostic transparency—factors that remain problematic in Bayesian and high-dimensional shrinkage approaches. The utility curvature coefficient $u$ introduces a quantifiable link between market structure and the utility of mean-variance optimization for a given asset set.

Practically, the model's adaptive mechanism and minimization of distance to the (forecasted) tangency portfolio offers robust performance in nonstationary and crisis regimes (notably during 2008 and 2020), suggesting applicability in institutional asset allocation, tactical rotation strategies, and context-sensitive risk management overlays.

Future Directions

The present analysis is limited to two asset universes and a modest number of predictive features. The paper suggests exploration of: (1) generalized universes for robustness analysis; (2) the impact of leverage limits and their interaction with portfolio risk; (3) enhancement of the VARX framework with technical indicators, macroeconomic variables, or modern automated selection methods (e.g., LASSO, nonlinear learners).

Notably, the paper claims that the approach is less prone to overfitting, more interpretable, and more robust than prior models forecasting Merton's coefficients, supporting further investigation into coefficient-based frontiers as a new paradigm for practical asset allocation.

Conclusion

This research advances portfolio selection methodologies by shifting the focus from direct estimation of the return-covariance pair to explicit forecasting of the efficient frontier's structure and leveraging minimum distance principles for robust portfolio construction. The approach achieves consistent and substantial improvements in realized Sharpe ratios, suggesting that efficient frontier forecasting and interpretable coefficient-based dimensionality reduction hold promise for future development in both the theoretical and applied domains of quantitative asset management (2604.03948).