Team Quantists — Suryansh Garg, Tanishq Godha, and Garv Jain — placed 87th globally 🌐(Top 0.7%) out of 12,620+ teams (20,000+ participants) in the IMC Trading Prosperity 3 Challenge 2025, and ranked 16th in India 🇮🇳.
This repository contains the codebase we developed throughout the competition. We hope it serves as a useful resource for anyone interested in quantitative finance, algorithmic trading, or competitive trading challenges.
| Round | Algo Score | Manual Score | Total Score | Rank |
|---|---|---|---|---|
| Round 1 | 0.000 | 44.340 | 44.340 | 2344 |
| Round 2 | 37.126 | 26.015 | 63.141 | 1180 |
| Round 3 | 234.523 | 51.500 | 286.023 | 194 |
| Round 4 | 137.892 | 61.002 | 198.894 | 82 |
| Round 5 | 12.084 | 154.505 | 166.589 | 87 |
| Overall | 421.625 | 337.362 | 758.987 | 87 |
Rainforest Resin was the simplest asset to trade since its price reliably oscillated around 10,000. We set initial thresholds to buy at ≤ 9,999 and sell at ≥ 10,001, then introduced a dynamic “edge” that, when orders at (9,999 buy, 10,001 sell) filled consistently, widened the band to buy at ≤ 9,998 and sell at ≥ 10,002, and, if fills became infrequent, narrowed it back to the original levels.
KELP followed a similar reversion idea, but with added filtering to avoid large (potentially toxic) orders. We calculated a fair price using filtered mid-prices and applied a basic reversion adjustment. The strategy had three parts: take good offers, clear risky positions, and place passive quotes near the fair value. This helped avoid bad fills and manage risk better.
For SQUID_INK, we used a stochastic oscillator to find possible trend reversals. Entry signals were based on %K and %D crossovers with basic price confirmation. Once in a trade, we tracked the highest/lowest price and used a 5% trailing stop to exit. This kept trades simple and avoided holding through big reversals.
In Round 2, we continued using the same base logic for RAINFOREST_RESIN, KELP, and SQUID_INK, with slight improvements. The main focus was on extending the strategy to support newly introduced assets: PICNIC_BASKET1, PICNIC_BASKET2, CROISSANTS, JAMS, and DJEMBES.
We introduced a basket arbitrage strategy for PICNIC_BASKET1 and PICNIC_BASKET2. The idea was to compute a theoretical price for each basket based on its components (e.g., CROISSANTS, JAMS, DJEMBES) and compare it with the live market price.
- If the basket was overpriced, we sold it and bought the components.
- If it was underpriced, we bought the basket and sold the components.
- All trades were hedged using component legs to keep exposure balanced.
- A dynamic threshold was used to account for volatility and avoid false signals.
These products were not traded independently. They were used exclusively for hedging the basket positions during arbitrage trades. Volumes were calculated based on the basket composition to neutralize exposure after a PICNIC_BASKET1 or 2 trade.
- RAINFOREST_RESIN and KELP: Slight tuning to edge widths and fair value smoothing.
- SQUID_INK: Refined reversion model and filtered price levels to reduce overtrading during noise.
Round 3 added new dimensions to the trading landscape, notably with VOLCANIC_ROCK and its VOUCHERS. We expanded our system while preserving all previous strategies.
We implemented a mean reversion strategy using Z-score logic over a rolling 50-tick window:
- Computed rolling mean and standard deviation of mid-prices.
- If Z-score < -1.9, we entered a long position.
- If Z-score > +1.9, we entered a short position.
- Position sizing was controlled using current exposure vs. soft limits.
This approach profited from temporary overreactions in price while managing risk via thresholds.
Each voucher (e.g., VOUCHER_9500, VOUCHER_9750, etc.) was handled using a strike-specific threshold for Z-score entry:
- When price deviated significantly (Z > threshold), a trade was executed in the opposite direction of the deviation.
- This strategy allowed for directional bets when mispricings occurred in relation to VOLCANIC_ROCK.
- Expanded inventory control and Z-score tracking.
- Added volatility-aware execution and smoothing on key indicators.
- Integrated minimal persistent state with
jsonpickleto retain history during backtesting.
Round 4 significantly expanded our trading system, introducing MAGNIFICENT_MACARONS, SUNLIGHT-based risk triggers, and more robust volatility-aware options strategies on VOLCANIC_ROCK VOUCHERS.
We developed a cross-market arbitrage strategy that used foreign market data and conversion costs to calculate implied bid/ask prices. Key innovations included:
- Edge adaptation: The quoting edge dynamically adjusted based on recent position volumes and market aggressiveness.
- Aggressive unwinding: If SUNLIGHT levels stayed below a critical threshold for multiple ticks, all MACARONS positions were aggressively cleared at market.
- Filtered execution: Only traded against favorable levels (filtered by size) to avoid adverse selection.
We introduced two volatility-driven strategies for options:
-
IV Momentum Strategy:
- Traded based on short vs. long EWMA IV crossover.
- Positioned long/short depending on the directional IV trend.
-
Arbitrage Strategy:
- Compared live mid-prices against Black-Scholes theoretical values with a dynamic buffer.
- Executed trades when prices deviated from theory, sizing based on vega and risk budget.
A meta-strategy switched between the two based on recent profitability.
To manage exogenous environmental risk, we:
- Monitored SUNLIGHT index for MACARONS.
- If readings stayed below 45 for 10 consecutive ticks, we immediately exited MACARONS positions.
- Prevented new entries until SUNLIGHT normalized.
- RAINFOREST_RESIN: No changes – base logic preserved.
- KELP: Improved quote filtering and smoother fair value adjustment.
- PICNIC_BASKETS: Enhanced dynamic thresholds and volume scaling for hedged arbitrage.
- SQUID_INK: Minor tuning of entry edges and reversion parameters.
- JAMS: New strategy using a stochastic oscillator to time reversion entries/exits.