Quantitative Trading (Quant) & High-Frequency Trading (HFT)

Here’s a breakdown of **quantitative and high-frequency trading (HFT) techniques** used by institutions, stripped down to their core mechanics:

—

### **1. Quantitative Trading (Quant)**
**Goal:** Use mathematical models to identify and exploit market inefficiencies.

#### **Key Techniques:**
– **Factor Models:** 
  – Analyze hundreds of factors (momentum, volatility, liquidity) 
  – Example: *Fama-French 3-Factor Model* extended for intraday signals. 

– **Machine Learning:** 
  – **Random Forests** for feature selection (which variables predict price moves). 
  – **LSTMs** to predict sequences (e.g., next 10 price ticks). 

– **Mean Reversion:** 
  – Trade assets that deviate from their statistical average (e.g., Bollinger Bands on steroids). 
  – Works best in range-bound markets. 

– **Kalman Filters:** 
  – Continuously update price relationships (e.g., for pairs trading). 

#### **Example:** 
A quant fund might use: 
“`math
Signal_t = 0.3*(Momentum) + 0.5*(Volatility_Adjusted_Mean_Reversion) + 0.2*(Liquidity_Score)
“`
Then trade when `Signal_t` crosses a threshold.

—

### **2. High-Frequency Trading (HFT)**
**Goal:** Profit from microsecond-level price discrepancies.

#### **Key Techniques:**
– **Latency Arbitrage:** 
  – Exploit price differences between exchanges (e.g., buy on NYSE at $50.00, sell on NASDAQ at $50.01). 
  – Requires **colocation** (placing servers inside exchange data centers). 

– **Order Book Dynamics:** 
  – **Sniping:** Detect large orders (via hidden liquidity patterns) and front-run them. 
  – **Spoofing:** Place fake orders to manipulate prices (now illegal, but still occurs in subtle forms). 

– **Market Making:** 
  – Quote bid/ask prices with a spread (e.g., buy at $99.99, sell at $100.01). 
  – Adjust quotes 1,000+ times per second to avoid adverse selection. 

– **Statistical Arbitrage (HFT Version):** 
  – Trade correlated ETFs (e.g., SPY vs. S&P 500 futures) with sub-second holding periods. 

#### **Example Workflow:** 
1. **Data Feed:** Raw order book updates (nanosecond timestamps). 
2. **Signal:** Detect a 0.005% mispricing between ES (S&P futures) and SPY. 
3. **Execution:** Send orders within **20 microseconds**. 
4. **Risk Check:** Automatically hedge with opposing futures trades. 

—

### **Tools & Infrastructure** 
| Component          | Purpose                                                                 |
|——————–|————————————————————————-|
| **FPGAs**          | Ultra-low latency hardware (faster than GPUs).                          |
| **Predictive APIs**| Forecast order fills before they happen (e.g., “Will my order execute?”).|
| **Clock Sync**     | Atomic clocks to sync timestamps across servers (critical for arbitrage).|

—

### **Real-World Example: Citadel’s HFT** 
– **Strategy:** Market making + statistical arbitrage. 
– **Speed:** ~15 microseconds to react to new data. 
– **Edge:** Pays for order flow from retail brokers (e.g., Robinhood) to predict price movements. 

—

### **Why Retail Traders Can’t Compete** 
– **Cost:** A single HFT server rack costs ~$500k/year. 
– **Data:** Institutional feeds show every order (not just candles). 
– **Speed:** Your broker’s 100ms latency vs. their 0.0001ms. 

—

### **What You Can Learn From Them** 
1. **Order Flow:** Tools like *BookMap* show hidden liquidity. 
2. **Algos:** Simple mean-reversion bots work on longer timeframes (1hr+). 
3. **Patterns:** Institutions often trade around key levels (e.g., options expiries). 

Want me to dive deeper into any specific area? (e.g., how quants backtest strategies or HFT’s “taker/maker” game).