Design Robinhood (Stock Trading Platform)

#Problem Statement

Design a retail stock trading platform where users can view real-time stock quotes, place buy/sell orders, and see their portfolio value. Orders must be routed to a market and confirmed back to the user.

📖

Real-world: The GameStop short squeeze (January 2021) exposed what's really under a trading app. As GME volume exploded, Robinhood abruptly restricted buying — not out of malice but because its clearinghouse (the NSCC) demanded a ~$3 billion collateral deposit overnight to cover the T+2 settlement risk of all those pending trades; Robinhood didn't have it and had to halt buys, drawing congressional hearings. The systems lesson: a trading platform is mostly a correctness-and-consistency problem, not a scale problem — orders are money, so you need a single-threaded matching/sequencing path per symbol, event sourcing for an immutable order ledger, idempotency to prevent double-execution, and strong consistency everywhere. "Eventually consistent" is acceptable for a feed; for someone's brokerage balance it's a lawsuit. Real-time quotes fan out over WebSockets, but the order path is where the rigor lives.

#Requirements

#Functional

• Real-time stock quotes (price, volume, bid/ask)
• Place market/limit/stop orders
• Order status updates (pending, filled, cancelled)
• Portfolio view (positions, P&L, history)
• Watchlist management

#Non-Functional

• Thousands of tickers, millions of users
• Quote latency: < 100ms (near real-time tick data)
• Order placement: < 500ms to confirmation
• Market hours: peak load 9:30am-4pm ET on trading days
• Regulatory: FINRA, SEC compliance; audit trail for every trade

#Core Design

#Real-Time Quote Data

Stock prices update multiple times per second during market hours. Source: market data feeds (NYSE, NASDAQ via OPRA, SIP). These provide raw tick data.

Market Data Feed → WebSocket receiver
                         ↓
                   Quote Normalizer (convert to unified format)
                         ↓
                   Kafka (topic: quotes, partitioned by ticker)
                         ↓
                   Quote Cache (Redis: latest price per ticker, TTL 1s)
                         ↓
                   Quote Service → SSE/WebSocket → User devices

For 5000 tickers × 100k active users subscribed to quotes: use pub/sub. Each quote server subscribes to Kafka, fans out to its connected clients.

Candles (OHLCV charts)1-minute aggregation from tick stream via Flink stream processor → stored in TimescaleDB.

#Order Routing

Robinhood is a broker-dealer. It doesn't operate its own exchange. Orders are routed to market makers (Citadel Securities, Virtu) or exchanges.

User places order → Order Service
                         ↓
                   Order Validation:
                   - Account balance / buying power check
                   - Position limits
                   - Pattern Day Trader checks
                         ↓
                   Order DB (write: PENDING)
                         ↓
                   FIX Protocol message → Market Maker / Exchange
                         ↓ (async fill notification)
                   Order fill webhook → Update Order DB (FILLED)
                   Update Portfolio DB
                   Notify user (WebSocket)

FIX ProtocolFinancial Information eXchange — the industry standard for order routing. Messages include order ID, ticker, quantity, side (buy/sell), order type.

#Portfolio Calculation

portfolio_positions(user_id, ticker, quantity, avg_cost_basis)
  → P&L = (current_price - avg_cost_basis) × quantity

Real-time portfolio value: quantity × current_quote (from Redis cache)
Historical P&L: store daily portfolio snapshots

#Architecture

Market Data Feed → Kafka → Quote Cache (Redis) → WebSocket → Users

Order Flow:
  User → Order Service → Validation → Order DB → FIX → Market
                                                ↓ (fill callback)
                                        Portfolio Service → User notification
                                        Trade History DB (append-only)

Compliance:
  Every order event → Audit Log (immutable, off-system) → FINRA reporting

#Key Design Decisions

#Security Considerations

#Interview Tips