Hi! I’m Shaman, a third-year CS undergrad from Mumbai who spends most of his time scrolling tech news on X. Last week my feed blew up with news that SEBI just slapped a ₹4,843 crore ban on U.S. quant giant Jane Street for “gaming” Nifty and Bank Nifty. Friends kept asking, “Bro, what exactly did their code do?” Here’s my beginner-friendly, ML-flavoured explainer.
TLDR
2023-25: Jane Street’s India desk pocketed ~₹36,000 crore trading weekly index options.
July 3 2025: SEBI orders an interim ban, calling their strategy Intraday Index Manipulation and Extended Marking the Close.
Core idea: use lightning-fast algorithms plus clever options maths to nudge the index a few points, then cash in on huge option positions whose payoff multiplies that tiny push.
They hid the stock-pushing leg inside an Indian subsidiary (legal for intraday) while the profit-making options sat with foreign FPIs (legal to hold overnight). Smart, but regulators said it crossed into manipulation.
The Algorithm Arsenal: Breaking Down the Tech Stack
1. The Hardware Foundation - Why Speed Was Everything
Jane Street's co-location setup at NSE was like parking your delivery bike inside the restaurant kitchen:
| Component | Purpose | Real-World Impact |
|---|---|---|
| FPGA-accelerated servers | Sub-20 µs order execution | 500x faster than retail traders |
| Dark fiber connections | Mumbai-Singapore-London real-time sync | Cross-market arbitrage in microseconds |
| Custom silicon NICs | Order-to-wire latency <200 ns | Hard-wired Black-Scholes calculations |
When you're competing for queue priority at NSE's matching engine, every microsecond is OP.
2. The Core AI Engine - Four Algorithms That Ruled Dalal Street
Algorithm 1: Gradient-Boosted Decision Trees (GBDT) - The Price Prophet
What it did: Micro-predicted whether Bank Nifty would tick up or down in the next 50 milliseconds.
How: Hundreds of tiny decision trees voting together on Level-2 order book data.
Why it mattered: Even 51% accuracy becomes gold when you trade 50,000 times per day.
# Simplified concept (not actual Jane Street code)
features = [
'bid_ask_spread',
'order_book_imbalance',
'volume_weighted_price',
'options_gamma_exposure'
]
prediction = gbdt_model.predict(features)
Algorithm 2: GRU-Based Sequence Model - The Pattern Hunter
Goal: Capture order-book sequences like "if bids stack fast, then price may pop."
Implementation: A GRU (lightweight LSTM cousin) processing time-series of market microstructure.
Edge: Combined recent momentum with longer context, crucial for intraday reversals.
Algorithm 3: Gamma-Scalping Engine - The Volatility Printer
This is where things get spicy. Jane Street allegedly ran huge negative gamma positions, then created the volatility to profit from them.
Simple analogy: They were short spring coils, then they shook the table so the coils would expand and pay out.
The Process:
- Morning (09:25-10:30): Buy ₹4,370 crore of Bank Nifty stocks hard → index jumps
- Afternoon (13:45-15:00): Aggressively dump same stocks → index crashes 1.8%
- Profit: Options explode in value during the whipsaw (₹735 crore gain vs ₹62 crore stock loss)
Algorithm 4: Option-Pinning Bot - The Settlement Sculptor
Goal: "Pin" expiry settlement exactly at strikes where Jane Street had maximum open interest.
Method: In the last 20 minutes, dump index futures in calculated micro-bursts.
Result: Retail OTM options expire worthless; Jane Street's book prints maximum profit.
The Technical Infrastructure Behind the Madness
As someone who's worked on distributed systems projects, I can appreciate the infrastructure needed:
Multi-Entity Coordination System
Jane Street didn't operate through a single account but coordinated activities across multiple entities simultaneously:
- JSI Investments (Indian subsidiary): Executed intraday stock legs
- Jane Street FPIs: Held overnight option positions
- Real-time synchronization: Microsecond-level coordination across entities
Surveillance Evasion Technology
Their order management system was designed to avoid detection:
- Fragmentation algorithms: Broke ₹4,000 crore trades into thousands of smaller orders
- Randomization techniques: Varied timing and sizing to appear natural
- Pattern testing: Likely ran simulations to ensure their activity wouldn't trigger alerts
Step-by-Step Breakdown of a "Manipulation Day"
Let me walk you through January 17, 2024 - one of the days SEBI flagged:
| Time Window | Action | Hidden Algorithm Logic |
|---|---|---|
| 09:25-10:30 | Buy ₹4,370 cr Bank Nifty stocks + futures (15-25% of volume) | Delta-neutral positioning: long stocks offset by short calls |
| 10:30-13:45 | Let market digest; retailers chase momentum | Gamma accumulation phase |
| 13:45-15:00 | Aggressive sell programs dump earlier longs | Trigger negative gamma explosion |
| 15:10-15:30 | Micro-burst futures sells every 5 seconds | Pin settlement at optimal strike |
Net result: ₹735 crore options profit minus ₹62 crore stock loss = ₹673 crore in one day.
The Machine Learning Insights That Fascinate Me
1. Feature Engineering Over Architecture
Jane Street's edge wasn't fancy transformers or GPT-style models. They won with:
- Order book imbalance ratios
- Real-time Greeks calculations
- Cross-asset momentum signals
- Liquidity heat maps
2. Latency as a Model Feature
Their GBDT wasn't just predicting prices - it was predicting prices given their execution speed advantage. The model knew it could act on predictions before others even saw the signal.
3. The Options Greeks Math
| Greek | Jane Street's Tactic | ML Component |
|---|---|---|
| Delta (Δ) | Stayed near-zero via futures hedging | Dynamic hedging ratios |
| Gamma (Γ) | Ran massive negative gamma positions | Volatility forecasting |
| Vanna | Sold high-vanna calls expiring worthless | Volatility surface modeling |
Why Their System Worked (And Why It Got Caught)
The Exploitation Framework
- Liquidity Asymmetry: Weekly options turnover (₹4 trillion/day) vs cash market (₹0.8 trillion/day)
- Regulatory Arbitrage: FPI intraday ban workaround using Indian subsidiary
- Speed Hierarchy: 500x latency advantage over retail participants
The Red Flags That Triggered SEBI
- Statistical impossibility: Correlation between cash and derivatives trades was too perfect
- Timing precision: Expiry day patterns suggested algorithmic coordination
- Scale inconsistency: Volume disproportionate to stated strategies
What This Case Teaches Us About Financial ML
The Technical Insights
- Speed + Statistics = Alpha: Model accuracy matters less than execution speed
- Feature Engineering Wins: Simple signals beat complex architectures when you have data advantage
- Infrastructure Is Everything: FPGA firmware matters more than Python notebooks
The Ethical Questions
As the next generation of ML engineers, we need to ask:
- How do we ensure our systems serve market integrity?
- What's the line between smart trading and manipulation?
- How do we balance innovation with ethics?
The Regulatory Response and Future Implications
SEBI's counter-moves show how regulators are adapting to algorithmic trading:
New Rules Implemented
- Delta-based limits: Net exposure monitoring vs just premium-based
- Group-level surveillance: Treating related entities as single books
- Real-time Greek monitoring: Gamma-alert dashboards at exchanges
What I'm Watching For
- ML-powered surveillance: How regulators will use AI to detect sophisticated strategies
- Industry response: Will quant firms self-regulate or push boundaries further?
- Global precedent: How other markets might adopt similar rules
My Takeaways as a CS Student
This case has fundamentally changed how I think about fintech ethics. Jane Street's team had incredible technical skills - the kind we all aspire to develop. But their alleged application raises crucial questions:
The Good
- Technical innovation: FPGA-accelerated ML in finance
- System complexity: Distributed real-time coordination
- Market efficiency: Sophisticated price discovery mechanisms
The Problematic
- Market integrity: Using tech advantages to potentially mislead others
- Regulatory evasion: Designing systems to operate in gray areas
- Fairness concerns: 500x speed advantages over retail participants
Looking Forward: Building Better Systems
As we develop our own algorithms, I believe we have a responsibility to:
- Design for transparency: Build systems that enhance rather than obscure market dynamics
- Consider all stakeholders: Not just profits, but market integrity and fairness
- Proactive compliance: Work with regulators rather than around them
- Ethical AI principles: Apply the same ethics we discuss in ML courses to finance
Final Thoughts
The Jane Street case represents the collision of cutting-edge technology with market regulation. Their algorithms demonstrated impressive technical sophistication, but according to SEBI, crossed into market manipulation territory.
The same technologies that can democratize financial services and create efficient markets can also be weaponized against market integrity. As the next generation of technologists, we have the power - and responsibility - to shape how these tools are used.
The question isn't whether we can build these systems. The question is: should we?
What are your thoughts on the Jane Street case? Are you working on fintech projects? I'd love to hear your perspectives. Follow me for more deep dives into the intersection of ML and finance.
Disclaimer: This analysis is based on publicly available information and SEBI's allegations. The case is ongoing, and Jane Street has not been convicted of any wrongdoing. This post is for educational purposes only.