•
Sen et al. (2023) compare mean-variance, hierarchical risk parity, and deep Q-learning portfolios on NIFTY50 and Indian sector stocks. Plain-language intro. Not investment advice.
Takeaway. Three allocation recipes on the same Indian-equity lab: mean-variance (MVO/MVP), hierarchical risk parity (HRP), and deep Q-learning. Score annual return, vol, Sharpe. The paper reports RL can look strong on Sharpe in their setup—treat that as a horse race on one market, not a law of nature.
A Comparative Analysis of Portfolio Optimization Using Mean-Variance, Hierarchical Risk Parity, and Reinforcement Learning Approaches on the Indian Stock Market (Jaydip Sen, Aditya Jaiswal, Anshuman Pathak, Atish Kumar Majee, Kushagra Kumar, Manas Kumar Sarkar, Soubhik Maji, 2023; arXiv:2305.17523).
Fit for learners: One paper, three allocation recipes on NIFTY50: Markowitz, hierarchical risk parity, and deep Q-learning—compare, don’t crown a winner for live cash.
This Hub note is a plain-language intro. Not investment advice.
After-work notebooks often pick one optimiser and overfit it. Markowitz needs means and a covariance (Ledoit-style shrinkage is the missing pre-step). HRP (López de Prado) clusters the corr matrix and allocates along a tree so you invert less junk. RL turns weights into an agent with a huge state space—hence a neural Q instead of a Q-table.
You can implement equal weight tonight, HRP-shaped clustering after you know linkage, and only then a DQN. The paper is a menu, not a production stack.
Tiny evening exercise: 4 fake assets; compute equal weights w = 1/n; then inverse-vol weights w ∝ 1/std. Print both. Learning only—no live orders.
Landing: arXiv abs
Not investment advice. This Hub note is educational only. Sen et al. is an academic comparison on historical Indian prices—not a recommendation to run MVO, HRP, or RL with real money.
Single course CTA: SoloQuant Academy’s Introduction to Python Programming—normalising a list of weights so they sum to one is a first-week loop.