Portfolio Optimization with Quadratic Transaction Costs

Institutional market impact modelling and portfolio optimization minimizing quadratic transaction costs

Results

	Mean-Variance	Market Impact	Alpha Model
Gross Return	14.3%	13.9%	6.2%
Net Return	-18.6%	-18.9%
Gross Information Ratio	0.797	0.802	0.870
Net Information Ratio	0.794	0.793
Avg. Turnover	2.47	2.47
Avg. Turnover – Optimized	-1.35	-1.37	1.39
Max Drawdown	42.2%	42.9%	11.9%
No. of Observations	11	11	95

Project Structure

portfolio_optimization/
┣ docs/                                     # Final report and slides
┃ ┣ report.pdf
┃ ┗ slides.pdf
┣ main/                                     # Model and backtest
┃ ┣ code.ipynb
┃ ┣ market_impact_cookbook.ipynb
┃ ┣ market_impact.py
┃ ┗ visualization.ipynb
┣ results/                                  # Results
┃ ┣ pf-daily-final.csv
┃ ┣ pf-result-final.csv
┃ ┣ img/
┃ ┣ ┣ Backtest-daily-ALL.png
┃ ┣ ┣ Backtest-daily-no-trading-costs.png
┗ ┗ ┗ Backtest-daily-trading-costs.png

• report is our final report, slides is our presentation slides.
• market_impact_cookbook is our documented code for the implementation of the Market Impact Model in Frazzini et al (2018), to estimate the transaction cost of any arbitrary trade
• market_impact is the module implementation of Frazzini et al (2018)
• code is our backtesting using alphas from non-linear factor modelling on US equities using RNNs, which also includes implementation for the Optimized Market Impact portfolio and the Optimized Mean-Variance portfolio described in the report.
• pf_daily-final and pf_results-final contains our backtest results.
• Macquarie Quant Alpha Model numbers are taken from Borghi & Giuliano (2020).

Credits

Our work could not have been possible without the portfolio optimization package PyPortfolioOpt