This project implements and backtests a long-only portfolio strategy based on a two-step procedure: in the first one, the algorithm screens the stock data in a rolling window to select a user-defined number of optimal ones, and, in the second, computes the portfolio allocation, using either Markowitz' mean-variance analysis or empirics. In this way, a potentially very large asset universe is shrunk to a few tens of assets, which dramatically reduces the estimation error of the covariance matrix of asset returns.
I perform the asset selection by recasting the Markowitz’ risk minimization problem as a regression problem, as explained in: Fan, J. and Zhang, J. and Yu, K. (2021). Vast Portfolio Selection with Gross-Exposure Constraints. Journal of the American Statistical Association, 107(498):592-606. This enables us to access the vast statistical toolbox associated to regression.
I build upon the approach presented in: Dan Wang, C. and Chen, Z. and Lian Y. and Chen M. (2020). Asset Selection based on High Frequency Sharpe Ratio. Journal of Econometrics, and I further enhance it by adopting the improved covariance matrix estimator, tested in the code in the ImprovingMVPortfolio repository.
The performance of the strategy is compared against the widely-used momentum strategy, its risk-managed counterpart and Dan Wang's method, which are all implemented in the code. The momentum strategy is presented in: Jegadeesh, N. and Titman, S. (1993). Returns to Buying Winners and Selling Losers: Implications for Stock Market Efficiency. Journal of Finance, 48:65–91. For risk-managed momentum, please refer to: Barroso, P. and Santa-Clara, P. (2015). Momentum Has Its Moments. Journal of Financial Economics, 116(1):111–120.
The dataset consists of daily closing prices and realized volatilities observed for U.S. stocks listed in the SP500 index in the time frame spanning between 1/1/2018 and 25/8/2020, including the market collapse due to COVID-19 pandemic.
They are located in the main() method in main.py
endow: (optional) Initial amount of money to be invested. Float
n_stk: Number of stocks to hold in the portfolio. Integer
n_obs_ar: Lengths of rolling window (in trading days). Array of integers
n_reb_ar: Portfolio rebalancing interval (in trading days). Array of integers
algo: Algorithm to use for stock selection. String. Possible choices are: 'mtm', 'rmmtm', 'sev' or 'sev+'
- mtm: Momentum. Pick stocks with largest value of average return
- rmmtm: Risk-managed Momentum. Pick stocks with largest value of average Sharpe ratio
- sev: Dan Wang's stock selection method
- sev+: Improved Dan Wang's method
wght_mtd: Stock weighting methods for portfolio allocation. List of strings. Possible choices are: 'equal', 'metric', 'mktcap', 'riskpar', 'lotp', 'tp'
- equal: Equally-weighted portfolio
- metric: Weights are proportional to momentum or SEV
- mktcap: Weights are proportional to market capitalization
- riskpar: Risk-parity portfolio
- tp: Tangency Portfolio (obtained via mean-variance analysis)
- lotp: Long-Only Tangency Portfolio (obtained via mean-variance analysis)
trsctn_fee_fix: (optional) Fixed transaction fees
trsctn_fee_prop: (optional) Proportional transaction fees