Smart Beta: Factor-Tilted Portfolio Construction

Smart beta is often discussed as a menu of factors (value, momentum, quality, low volatility). In practice, two decisions matter: which systematic risks you want to hold, and how you translate that view into position sizes. This study holds the investable universe fixed and changes only the construction rule, so differences in outcomes can be read as consequences of portfolio engineering rather than stock picking.

The goal is not to declare a single winning sleeve, but to show how benchmark choice, risk budgeting, and deliberate factor tilts interact over a recent one-year window. That framing helps answer whether a style bet failed because the factor was weak, because the portfolio was concentrated, or because diversification across construction approaches would have mattered.

The investable set is a broad U.S. large-cap panel balanced across industry sectors, so no single sector dominates the baseline. Performance is evaluated over the most recent 252 trading days (approximately one calendar year of market days). All rolling estimates of risk, factor loadings, and covariances use the same one-year lookback, so every sleeve is judged with consistent information timing.

Daily factor returns follow a standard six-factor panel: market, size, value, profitability, investment, and momentum. Stock returns are aligned to those dates so portfolio exposures can be interpreted in the same language as academic asset-pricing studies.

For each stock $i$ and day $t$, excess return is approximated by a linear factor structure:

$r_{i,t}-r_{f,t}={\alpha }_{i,t}+{\sum }_k{\beta }_{i,t}^{(k)}{f}_{k,t}+{\epsilon }_{i,t}$

where $f_{k,t}$ are factor returns (market, size, value, profitability, investment, momentum) and ${\beta }_{i,t}^{(k)}$ are estimated from a trailing year of daily data. Portfolio factor exposure on day $t$ is the weighted sum ${\sum }_i{w}_{i,t-1}{\beta }_{i,t}^{(k)}$: holdings weights from the prior close map stock-level loadings into sleeve-level risk.

All sleeves are long-only and fully invested (${\sum }_i{w}_{i,t}=1$, $w_{i,t}\ge 0$) and rebalance daily. Weights applied at $t$ are those chosen at $t-1$, so reported returns do not use same-day information.

Equal-weight benchmark: $w_{i,t}=1/N$ for all names in the panel.

Minimum-variance sleeve: choose weights that minimize portfolio variance ${\sigma }_p^2={\mathbf{w}}^{\prime }{\Sigma }_t\mathbf{w}$ subject to the budget and long-only constraints, where ${\Sigma }_t$ is the trailing one-year covariance matrix (with mild shrinkage for stability).

Value-tilt sleeve: overweight names with high trailing HML loading. Operationally, names in the upper part of the cross-sectional HML beta distribution receive equal weight within that selected group.

Value–momentum blend: rank names by a combined score $0.5\cdot \text{rank}({\beta }^{HML})+0.5\cdot \text{rank}({\beta }^{MOM})$ and hold an equal-weight subset of top-ranked names, aiming to balance two premia that often diverge through time.

The results section presents a full dashboard (no hidden views): performance and risk table (return, volatility, Sharpe ratio, maximum drawdown, win rate, turnover), cumulative wealth paths, drawdown paths, rolling annualized volatility, factor return correlation and cumulative factor performance over the study window, average and time-series factor exposures for each sleeve, sector allocation at the end of the window, and correlation of daily portfolio returns across sleeves.

Together these views separate return generation, risk budgeting, style exposure, sector concentration, and co-movement between construction approaches.

Figures and tables use the latest available market data through the date shown in the results banner. Refresh the study periodically if you want the one-year window to stay current with live prices and factor updates.