Round 1 winner · trading agents
Arnav's winning algorithm.
Arnav won Round 1 with a concentrated leader-momentum bot that pressed winners when the market regime was clean, added a measured 2x ETF sleeve only in the strongest state, and cut exposure hard when the tape broke. This is the benchmark Round 2 entrants need to beat.
What It Was
The submission was not a buy-and-hold basket and it was not an all-in leveraged bet. It was a state machine with three modes: CASH, NEUTRAL, and FULL. The bot only used the bars provided by the challenge engine, made deterministic decisions, and kept every position inside the concentration and leverage caps. The actual submitted Python file is public here, so competitors can study the code path directly instead of guessing from the leaderboard.
Core Algorithm
- Rank market leaders: score large-cap tech, AI/chip names, broad ETFs, and sector ETFs using 42-day momentum, 21-day momentum, and distance above the 50-day moving average.
- Hold the top five: select only names with positive momentum and price above their 50-day average, then weight them by rank so the strongest names get more capital.
- Change exposure by regime: go FULL only when SPY and QQQ are above fast and slow trend filters, breadth is healthy, and QQQ volatility is contained. Otherwise run NEUTRAL or CASH.
- Use leverage carefully:QLD and SSO appear only in FULL mode, never as a standing bet. The full book is clamped below the challenge's beta-adjusted gross cap.
Risk Controls
Hard cash state
If SPY or QQQ breaks below the slow trend band, or if QQQ has a fast crash/volatility brake, the bot exits risk.
Hysteresis
After going to CASH, the bot waits for a clear reclaim instead of instantly rebuying on noise.
Drawdown taper
At roughly -6% drawdown it cuts target exposure in half; around -10% it locks down to about one-quarter exposure.
Stops and cooldowns
An 8% trailing stop exits individual positions and blocks immediate re-entry for a few days.
Position caps
Targets stay below 26% per name, with drift trimming around 28%, under the 30% rule.
Rebalance discipline
The bot normally rebalances every three trading days, but derisks sooner on regime breaks or drift.
Why It Worked
Round 1 rewarded the combination of direction and restraint. The market favored leaders, so Arnav's bot owned the right kind of assets. But the key difference was that it did not keep pressing risk blindly: it had a market-wide brake, a drawdown taper, per-position stops, and caps that prevented one name or one leveraged sleeve from dominating the result.
How To Beat It
Round 2 entrants do not need to copy this. They need to improve one of the tradeoffs: faster regime detection, better breadth signals, cleaner position sizing, stronger drawdown recovery, or a differentiated universe that still respects the caps.
A good submission should answer one question clearly: when should the bot be aggressive, and what exactly forces it to stop being aggressive?
Educational challenge notes only. This is not investment advice, and it is not a recommendation to trade any security.