Optimize Portfolios with
Quant-Grade Precision

Portfolio optimization is the process of selecting the best asset weights to maximize returns or minimize risk for a basket of Indian equities. Folio Lab applies quantitative methods like Mean-Variance, Hierarchical Risk Parity, and Black-Litterman to NSE and BSE stocks, computing optimal weights backed by decades of academic research.

There is no single best method - it depends on your goals. Mean-Variance works well with reliable return estimates, Risk Parity is robust when you want balanced risk exposure, and HRP excels when dealing with correlated Indian market sectors. Folio Lab lets you compare all 30 methods side-by-side.

Manual allocation relies on intuition and simple heuristics. Folio Lab uses mathematically optimal algorithms that account for correlations, volatility, and tail risk across your entire portfolio - delivering measurably better risk-adjusted returns as shown by decades of academic research.

Folio Lab is currently in beta and offers a free tier that lets you run optimizations, analyze results, and access the full documentation. Depending on how this beta performs, a full version with retail and enterprise pricing will be released. Create an account to get started in minutes.

Folio Lab calculates 50+ metrics including Sharpe Ratio, Sortino Ratio, Value at Risk (VaR), CVaR, Maximum Drawdown, Jensen's Alpha, Treynor Ratio, Information Ratio, and multiple beta variants - all computed specifically for your optimized Indian equity portfolio.

Yes. You can optimize portfolios on either exchange - choose NSE stocks with Nifty or Bank Nifty as the benchmark, or BSE stocks with Sensex. Each optimization run uses a single exchange for consistent pricing and benchmark comparison.

Yes. Pro and Enterprise plans support US-listed equities (NYSE / NASDAQ), benchmarked against the S&P 500, Nasdaq-100 or Russell 3000, with US Treasury or fed-funds risk-free rates from FRED. US and Indian assets can't be mixed in a single run.

Folio Lab can optimize portfolios with hundreds of Indian stocks simultaneously. The platform handles large covariance matrices efficiently, even for methods like Mean-Variance and Black-Litterman that require matrix inversion.

Folio Lab uses historical price data for NSE and BSE equities, Indian government bond yields for risk-free rates, and Nifty 50 / Sensex as benchmark indices. All data is sourced and processed for accurate covariance estimation and return calculation.

An LLM predicts plausible-looking text - it does not run a solver. Ask it for max-Sharpe weights on 30 stocks and it will invent numbers that look reasonable but do not sit on the efficient frontier and do not satisfy your constraints. Folio Lab runs an actual convex optimizer (quadratic and conic programming) that returns the true mathematical optimum to numerical tolerance, every time, for the same inputs.

No. Risk optimization lives and dies on the covariance matrix. For 30 assets that is 435 unique pairwise covariances, each estimated from price history, shrunk (Ledoit-Wolf) and checked to be positive semi-definite before the matrix is inverted. A language model cannot hold those numbers, let alone invert the matrix - it approximates, and a wrong covariance matrix produces confidently wrong weights. Folio Lab computes it numerically from real price series.

No. An LLM has already seen the outcomes in its training data, so its 'backtest' is narration of what worked - lookahead bias by construction. It also ignores transaction costs, slippage, STT and brokerage, lot-size rounding and rebalancing schedules, so its returns are fiction. Folio Lab enforces a strict point-in-time cursor and models real-world frictions, so the numbers survive due diligence.

If you manage other people's money you are a fiduciary. When a SEBI auditor or an allocator asks how a number was produced, you must reproduce it bit-for-bit from a versioned config and dataset. The same prompt to an LLM gives a different answer next week - that alone disqualifies it for regulated, audited use. Folio Lab runs are versioned, seeded and reproducible by design.

No. A language model has stale, partial training text - it 'knows' only the companies that survived to be written about and has no point-in-time, split- and dividend-adjusted NSE/BSE series, and no delisted names. Backtests on that data overstate returns. Folio Lab uses adjusted, point-in-time price data that includes dead tickers so risk and return are measured honestly.