Doohwi Cha
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Doohwi Cha has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Yoontae Hwang
Pusan National University, Busan, Republic of Korea
7 shared publications - Donghwa Seo
DS Investment & Securities, Seoul, Republic of Korea
6 shared publications - Juyeong Lee
EY Consulting, Seoul, Republic of Korea
6 shared publications - Minjae Lee
Independent Researcher, Seoul, Republic of Korea
6 shared publications - Minsuk Sung
Korea University, Seoul, Republic of Korea
6 shared publications - Seunghan Son
Independent Researcher, Seoul, Republic of Korea
6 shared publications
Latest Publications
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 35-43
AlphaLeak: What `Blind' LLM Trading Benchmarks Still See
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 35-43
Open AccessAlphaLeak: What `Blind' LLM Trading Benchmarks Still See
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1348-1356
Neural Estimation of Irreversibility in Real and Simulated Limit Order Book Paths
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1348-1356
Open AccessNeural Estimation of Irreversibility in Real and Simulated Limit Order Book Paths
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1110-1118
Metropolitan Housing Signals for Treasury Duration Risk Management
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1110-1118
Open AccessMetropolitan Housing Signals for Treasury Duration Risk Management
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 568-576
SlipClock: A Conditional Displayed-Depth Cost Benchmark for Evaluating Financial AI Strategies in Crypto Futures
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 568-576
Open AccessSlipClock: A Conditional Displayed-Depth Cost Benchmark for Evaluating Financial AI Strategies in Crypto Futures
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 67-76
The Division of Research: Information Access and Independent-Book Capacity in LLM Investment Teams
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 67-76
Open AccessThe Division of Research: Information Access and Independent-Book Capacity in LLM Investment Teams
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1452-1460
Decision-Focused Learning of the Gerber Threshold
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 1452-1460
Open AccessDecision-Focused Learning of the Gerber Threshold
- Open Access
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 940-948
CallRank: Isolating What Changed in Earnings-Call Q&A for Sector Ranking and Cost-Adjusted Alpha
- 13 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 940-948
Open AccessCallRank: Isolating What Changed in Earnings-Call Q&A for Sector Ranking and Cost-Adjusted Alpha
- Open Access
- 14 July 2026
- Published by Elsevier in SSRN Electronic Journal
Is Compute the New Oil? Cross-Asset Risk and the Case for a Dedicated Market
GPU compute has become a multi-hundred-billion-dollar exposure underpinning AI, yet whether its price risk can be hedged with existing assets—and thus whether the dedicated compute-futures markets announced in 2026 are warranted—has, to our knowledge, not been tested. Using daily GPU rental-rate benchmark indices (via Bloomberg) for three generations (A100, H100, B200)—the first such cross-generation panel we are aware of—we ask whether compute is "the new oil": a hedgeable industrial commodity. We document its price dynamics—newer generations are far more jump-prone than oil or equities (though lower in overall volatility), with no volatility clustering—and a cross-generation price structure whose discounts shift over time. We then test cross-asset proxy hedging out-of-sample. GPU rental returns are weakly correlated (daily |ρ| ≲ 0.1) with NVIDIA, semiconductors, compute-infrastructure equities, and the broad market, and no minimum-variance proxy hedge delivers out-of-sample variance reduction distinguishable from zero—across the three generations, across horizons from daily to weekly (monthly and quarterly results are indicative only, given few non-overlapping blocks), and after multiple-testing, active-day (stale-filtered), and frontier-roll checks; a naïve one-for-one hedge sharply adds risk. This is a negative result on a short, stale sample: a power analysis shows the effective sample cannot resolve a true variance reduction below roughly 9%, so we report the absence of a detectable conventional hedge rather than proof of exact orthogonality. Even so, no detected hedge removes more than a small fraction of a jump-prone exposure, so the practical case for a direct instrument is little changed by that ceiling. We read the result constructively—a suggestive incomplete-market rationale for a dedicated market—while noting that the same weak correlation implies a liquidity paradox for those contracts, and we quantify the heightened exposures (not materially reducible by the proxies we test) in AI-training budgets and GPU-collateralized lending.
- 14 July 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessIs Compute the New Oil? Cross-Asset Risk and the Case for a Dedicated Market
GPU compute has become a multi-hundred-billion-dollar exposure underpinning AI, yet whether its price risk can be hedged with existing assets—and thus whether the dedicated compute-futures markets announced in 2026 are warranted—has, to our knowledge, not been tested. Using daily GPU rental-rate benchmark indices (via Bloomberg) for three generations (A100, H100, B200)—the first such cross-generation panel we are aware of—we ask whether compute is "the new oil": a hedgeable industrial commodity. We document its price dynamics—newer generations are far more jump-prone than oil or equities (though lower in overall volatility), with no volatility clustering—and a cross-generation price structure whose discounts shift over time. We then test cross-asset proxy hedging out-of-sample. GPU rental returns are weakly correlated (daily |ρ| ≲ 0.1) with NVIDIA, semiconductors, compute-infrastructure equities, and the broad market, and no minimum-variance proxy hedge delivers out-of-sample variance reduction distinguishable from zero—across the three generations, across horizons from daily to weekly (monthly and quarterly results are indicative only, given few non-overlapping blocks), and after multiple-testing, active-day (stale-filtered), and frontier-roll checks; a naïve one-for-one hedge sharply adds risk. This is a negative result on a short, stale sample: a power analysis shows the effective sample cannot resolve a true variance reduction below roughly 9%, so we report the absence of a detectable conventional hedge rather than proof of exact orthogonality. Even so, no detected hedge removes more than a small fraction of a jump-prone exposure, so the practical case for a direct instrument is little changed by that ceiling. We read the result constructively—a suggestive incomplete-market rationale for a dedicated market—while noting that the same weak correlation implies a liquidity paradox for those contracts, and we quantify the heightened exposures (not materially reducible by the proxies we test) in AI-training budgets and GPU-collateralized lending.