The Conditional Value of High-Frequency Covariance Estimates in Minimum-Variance Portfolio Optimization: Evidence from Borsa Istanbul

This study examines the conditional value of high-frequency covariance estimates for minimum-variance portfolio construction in Borsa Istanbul. Using Matriks adjusted 5-minute OHLC data for 11 liquid stocks that were BIST 30 constituents at the beginning of 2017 and have continuous observations over 2017-2025, realized covariance matrices are constructed at 5-minute and 30-minute frequencies. The portfolios are compared with daily-covariance MVPs, an Equal-Weighted benchmark, a buy-and-hold portfolio, and the BIST 30 Total Return Index. The out-of-sample analysis covers 2018-2025 and considers daily and monthly rebalancing, four rolling windows, no short selling, a 20% stock-level cap, and transaction costs of 0.10% for optimized strategies. USD-denominated and inflation-adjusted returns are also examined. Results show that high-frequency covariance estimates do not provide unconditional dominance. Their contribution is concentrated under the 3-month window, where they produce competitive risk-adjusted performance and lower turnover than daily-covariance MVPs at similar volatility levels. The findings suggest that the practical value of high-frequency covariance depends on sampling frequency, estimation window, rebalancing frequency, transaction costs, and benchmark choice.