Risk management framework
The signal is one layer; the risk-control layer is what turns it into a tradeable book. Below are the hard limits a real desk would impose, and a backtest review showing how the existing trade history would have respected (or breached) them.
Defined limits
Per-issuer notional cap
¥200,000,000 hard ceiling per issuer regardless of how cheap. Stops the strategy from over-concentrating in a single name even if the signal stays strong.
Per-issuer concentration
No single issuer can exceed 25.0% of equity at any time. The backtest concentration analysis showed our top 5 issuers were 113% of P&L — this limit forces broader diversification.
Per-trade size cap
Maximum 20.0% of equity in any single trade. Matches the 5-slot sizing scheme.
VaR limits
1-day 95% historical VaR: warn at 2.0% of NAV (soft), block new entries at 4.0% (hard). Standard hedge-fund discipline.
Drawdown circuit breakers
At -5.0% rolling drawdown: halve all new position sizes. At -10.0%: halt new entries entirely.
Total deployment
Max 100.0% of equity deployed simultaneously. No leverage on top of cash; cash earns the JPY money-market rate.
Backtest review · how the limits would have behaved
Trades that hit a limit
| Entry date | Issuer | Entry cheap% | Proposed ¥ | Accepted ¥ | % equity | Decision | Why |
|---|---|---|---|---|---|---|---|
| 2023-06-16 | Mercari | +19.3 | 150,000,000 | 37,500,000 | 5.0% | RESIZED | issuer_notional_cap (cur=150M + 150M > 200M); issuer_pct_cap (26.7% > 25.0%) |
| 2024-08-02 | Daifuku | +6.5 | 148,812,994 | 37,203,248 | 5.0% | RESIZED | issuer_notional_cap (cur=149M + 149M > 200M); issuer_pct_cap (26.9% > 25.0%) |
| 2024-08-05 | Nagoya Railroad | +5.2 | 148,812,994 | 37,264,406 | 5.0% | RESIZED | issuer_notional_cap (cur=149M + 149M > 200M); issuer_pct_cap (26.9% > 25.0%) |
| 2024-08-05 | Kobe Steel | +8.4 | 148,812,994 | 37,203,248 | 5.0% | RESIZED | issuer_notional_cap (cur=149M + 149M > 200M); issuer_pct_cap (26.9% > 25.0%) |
| 2024-11-01 | Daifuku | +7.1 | 149,627,509 | 37,406,877 | 5.0% | RESIZED | issuer_notional_cap (cur=150M + 150M > 200M); issuer_pct_cap (26.7% > 25.0%) |
| 2024-11-04 | Kobe Steel | +5.2 | 149,627,509 | 37,406,877 | 5.0% | RESIZED | issuer_notional_cap (cur=150M + 150M > 200M); issuer_pct_cap (26.7% > 25.0%) |
| 2025-04-07 | Daifuku | +6.3 | 145,966,746 | 36,491,686 | 5.0% | RESIZED | issuer_notional_cap (cur=146M + 146M > 200M); issuer_pct_cap (27.4% > 25.0%) |
| 2026-03-06 | Rohm | +5.3 | 144,964,969 | 36,241,242 | 5.0% | RESIZED | issuer_notional_cap (cur=145M + 145M > 200M); issuer_pct_cap (27.6% > 25.0%) |
| 2026-04-22 | Daifuku | +5.0 | 144,964,969 | 36,241,242 | 5.0% | RESIZED | issuer_notional_cap (cur=145M + 145M > 200M); issuer_pct_cap (27.6% > 25.0%) |
What's not modelled (be honest in interviews)
- Total book vega cap is defined but not yet enforced trade-by-trade in the simulator — would need per-trade vega and a running sum.
- Stress tests (parallel rate shifts, equity crash, vol spike) — not implemented; a real desk runs nightly stress.
- Counterparty / funding limits — assumes unlimited repo and stock-borrow availability, which isn't realistic for size.
- Operational limits — bond settlement (T+2), corporate action handling, ex-coupon dates — abstracted away.
- Correlation-aware concentration — two unrelated issuers in the same sector still co-move; a real limit would group by sector beta.