Quantum-aware algorithmic trading platform for prop desks, HNI portfolios and quant funds.
QuantEdge is a multi-asset execution and signal platform for prop desks, HNI portfolios and quant funds. The same adversarial-AI core that hardens DRISHTI-AKSHI runs the alpha models — trained against synthetic market regimes, stress-tested with QAOA portfolio search.
US 50-asset universe · 2018-12-24 → 2025-12-30 · weekly rebalance · net of 10bps costs. Walk-forward, no lookahead. Our adversarial-robust H2PSRO and neural ARL allocators are benchmarked against classical Mean-Variance and Equal-Weight — the two baselines every institutional allocator already knows.
NOTE · Curves are a deterministic reconstruction from the reported summary statistics (CAGR, vol) with historical regime shocks (COVID, 2022, SVB) layered per strategy beta. Terminal equity matches reported CAGR exactly; intra-period path is illustrative. Full weekly returns available under NDA.
| Metric | Equal Weight | Mean-Variance | ARL (Neural) | H2PSRO (Robust) |
|---|---|---|---|---|
| CAGR | 15.59% | 20.59% | 13.39% | 14.12% |
| Volatility | 16.58% | 16.52% | 17.38% | 14.95% |
| Sharpe Ratio | 0.837 | 1.095 | 0.696 | 0.825 |
| Max Drawdown | −30.43% | −23.58% | −33.35% | −28.51% |
| Calmar Ratio | 0.512 | 0.873 | 0.402 | 0.495 |
| Ann. Turnover | 0.00× | 6.28× | 0.09× | 0.50× |
Static adversarial-trained weights. Sharpe 0.825 at just 14.95% vol — the lowest of any strategy — with only 0.50×/yr turnover. Value/defensive tilt: Energy, Healthcare, Fixed Income.
Learned allocator with near-zero turnover (0.09×/yr). Diversification benefit vs classical MV — low return correlation makes ARL/H2PSRO complementary sleeves inside a blended book.
SOURCE: Internal walk-forward report · scipy.optimize (MV) · numpy (EW) · ParaMedha stack (ARL, H2PSRO) · full report available under NDA.
1. Our robust allocator is real, not a demo. H2PSRO delivered a 0.825 Sharpe at the lowest volatility of the four strategies and only 0.50×/yr turnover — competitive with classical Mean-Variance while surviving adversarial stress it was explicitly trained against.
2. Mean-Variance is a strong benchmark, not a ceiling. MV won headline CAGR (20.6%) and Sharpe (1.09), but at 6.28×/yr turnover — every rebalance eats costs, slippage and capacity. That number is what stops MV from scaling in real books.
3. ARL is a diversifier, not a replacement. The neural allocator's returns are only weakly correlated with MV — meaning a blended ARL + MV sleeve improves risk-adjusted returns beyond either alone. That is the sleeve QuantEdge ships to clients.
4. Everything is walk-forward. No lookahead, no cherry-picked windows, no in-sample flattery. The same code runs the live book.
A naive 2% allocation to each of the 50 assets, held for the entire 6.5-year window. No signals, no rebalance, no view. This is the 'do-nothing' benchmark every quant strategy must beat.
If a sophisticated model cannot beat equal-weight after costs, it has no reason to exist. EW is the honest zero-line — no data-mining, no parameter risk.
CAGR 15.59% and Sharpe 0.837 set the floor. Anything below this line is destroying value. Note the −30.43% drawdown: 'safe and simple' still bleeds in a real bear.
Classical mean-variance optimisation solved with scipy.optimize on a rolling 126-day covariance and returns estimate. Long-only, no leverage, weekly rebalance, net of 10bps costs.
MV is the textbook — the strategy every allocator, endowment and pension desk already runs. If we cannot cite it side-by-side, our numbers are unverifiable.
Best headline numbers: CAGR 20.59%, Sharpe 1.095, Calmar 0.873. But turnover is 6.28×/yr — the entire book flips six times a year. In real money that hits capacity, tax and slippage hard.
A reinforcement-learning agent trained on the same 50-asset universe, rewarded for long-horizon compounding under adversarial market perturbations. Rolled forward on live windows.
This is our proof that a neural allocator can learn genuinely different behaviour from MV — not just a noisier copy. Low correlation to MV is the whole point.
Modest CAGR (13.39%) and Sharpe (0.696) on its own, but turnover is 0.09×/yr — effectively buy-and-adjust. Its value shows up when blended with MV, not on its own line.
A robust portfolio solved via Hamiltonian Policy-Space Response Oracle: worst-case adversary picks the return distribution, allocator responds, iterate to equilibrium. Static weights, held forward.
MV assumes tomorrow looks like the last 126 days. H2PSRO explicitly assumes it will not. This is the sleeve that has to survive regime breaks — 2020, 2022, and whatever comes next.
Sharpe 0.825 at just 14.95% volatility — the lowest of any strategy. Only 0.50×/yr turnover, smallest drawdown among neural approaches. Trades peak CAGR for a smoother, defendable ride.
HOVER ANY DOTTED TERM ABOVE FOR ITS EXACT FORMULA · OR JUMP TO THE FULL GLOSSARY.
Datasheets are issued to verified defence, government and integrator contacts only. Submissions are logged and reviewed by the ParaMedha programme office.
ParaMedha leads as prime integrator and AI/fusion IP owner. Consortium partners include BEL, SAMEER, Tonbo Imaging, IIT Madras CAI, Bharat Dynamics and DRDO labs.
PhD, UTS Sydney · Assistant Professor, BITS Pilani. Three Springer monographs, ten peer-reviewed papers on adversarial machine learning.
Indigenous. Aatmanirbhar Bharat. Built for the next Operation Sindoor.