This repository benchmarks four tail-risk optimization objectives (CVaR, spectral, distributionally robust CVaR, entropic VaR) against hard-coded threshold rules for protecting a diversified short book from intraday squeezes in Binance USD-M perpetual futures, finding that optimizer structure and recovery policy matter far more than the choice of risk functional.
13
No data
2
0
MIT
2026-07-29
The study provides a fully causal, reproducible pipeline with open code and data, revealing that the optimizer's recovery path dominates risk functional choice, and that tuning parameters on pre-incident data collapses protection for the fastest squeezes—a critical warning for quantitative risk management.
A study report benchmarking exposure-control objectives (CVaR, spectral risk measures, distributionally robust CVaR, entropic value-at-risk) against a hard-coded threshold rule for protecting a diversified short book from parabolic intraday moves in Binance USD-M perpetual futures. The motivating discussion is preserved with full attribution in SOURCE.md.
Note: every policy is scored on the same three axes: protection (fraction of peak squeeze loss avoided), cost (execution and turnover drag), and utility (net result once cost is set against protection). The universally optimal hedging design remains open. The full pipeline is published so every number can be checked; corrections are welcome.
Background. A short book in crypto perpetuals carries unbounded right-tail risk. A small-cap contract can multiply intraday on no news, and a daily rebalance leaves the original short size exposed for hours; the July 26 PIEVERSE pump, during which realized volatility reached twice its forecast within ten minutes, is the motivating incident. Open design issues include the choice of intraday lookback, the turnover cost of reacting quickly, the documented upward bias of five-minute volatility forecasts, and the rule by which falling volatility should rebuild a reduced position.
Methods. The sample covers all 697 Binance USD-M perpetuals with complete history from May through July 2026, resampled to 5-minute bars. A two-horizon volatility-acceleration score (100-minute against 48-hour RMS volatility, computed on BTC-residualized returns) triggers either a hysteretic threshold overlay or one of four tail-risk optimizers over drift-conditioned scenarios: CVaR, a spectral Kusuoka mixture, Wasserstein distributionally robust CVaR, and entropic value-at-risk. Execution is causal (next bar) and charged 20 bps one-way. June selects the lookback; July 1–8 calibrates thresholds; July 8–22 is the untouched test window; July 22–28 contains the verified incidents.
Results. The 20-bar window maximizes forecast rank correlation with next-25-minute realized volatility (median pair-level Spearman 0.49 in selection, 0.41 out of sample). Threshold rules protect 70.3–73.9% of aggregate peak episode loss where price rose at least 10% after t
Similar projects matched by category, topics, and programming language.
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.
mimic captures traffic from any iOS or web app and automatically generates a Python client library that lets you call the app's API like a regular library.