BTC Market Microstructure — Volatility, Liquidity & Market Resilience

Market Microstructure · Realized Volatility · Liquidity Analysis · High-Frequency Data · R

FE-570 Final Project — MS Financial Engineering, Stevens Institute of Technology

Author: Swara Dave

---

📌 Overview

This project analyzes intraday volatility, liquidity, and market resilience in cryptocurrency markets using tick-level BTC/USD trading data from BitMex. The study investigates how market participants respond to volatility through liquidity measures and how large trades impact prices during volatile periods.

Key questions addressed:

1. How can realized volatility be estimated from high-frequency tick data?
2. How do liquidity spreads behave during high vs. normal volatility periods?
3. What is the price impact of large trades (>40,000 units) on market resilience?

---

📊 Key Results

Volatility Estimates

Method	Annualized Volatility
5-Minute Realized Volatility	30.76%
Roll Model Volatility	28.97%

Liquidity Spread Analysis

Spread Measure	Mean (USD)	Std Dev	Annualized
Quoted Spread	0.569	0.816	9.029
Effective Spread	0.406	0.572	6.438
Roll Spread	0.0000227	0.0000848	0.00036

Key finding: Quoted spread significantly exceeds effective spread, indicating traders often execute at prices better than displayed quotes — evidence of hidden liquidity and strong price discovery in BTC markets.

---

🗂️ Data

• Source: BitMex trading API (Python)
• Asset: BTC/USD (XBTUSD)
• Period: April 17, 2017 — full 24-hour trading day
• Granularity: Tick-level (millisecond timestamps)
• Size: 14,618 observations
• Fields: timestamp, trade price, bid price, ask price, volume, trade side

---

⚙️ Methodology

1. Volatility Estimation

Realized Volatility (Signature Plot)

• Computed realized variance RV(q) = (1/q) Σ(P_{t+q} - P_t)² for lags q = 1 to 500
• 5-minute realized volatility: 26.13% → annualized to 30.76%
• Signature plot shows volatility decreasing as lag increases — confirming microstructure noise at small lags

Roll Model Volatility

• Estimates true bid-ask spread from serial covariance of price changes
• σ²u = γ0 + 2γ1 → annualized to 28.97%
• Close agreement between both methods confirms robustness

2. Liquidity Analysis

Three spread metrics computed:

• Quoted Spread = Ask - Bid (posted liquidity cost)
• Effective Spread = 2 × D_t × (Price - Midpoint) (actual execution cost)
• Roll Spread = 2 × √(-Cov) (true cost after removing microstructure noise)

Spreads analyzed separately during high volatility vs normal periods using rolling 20-trade window.

3. Large Trade Impact Analysis

• Large trades defined as volume > 40,000 units
• Immediate Price Impact: mostly contained between -0.1% and 0.2% — good market resilience
• Cumulative Price Impact (10-tick window): wider dispersion (-0.4% to 0.2%) with slight downward trend
• Increasing dispersion in latter half of sample suggests decreasing resilience over the trading day

---

📁 Repository Structure

BTC-Market-Microstructure/
├── analysis.R       # Volatility estimation, liquidity analysis, large trade impact
└── README.md

---

🚀 How to Run

1. Clone the repo
2. Install required R packages:

install.packages(c("xts", "highfrequency", "data.table", "dplyr", "ggplot2", "zoo", "tidyr"))

3. Obtain BTC/USD tick data from BitMex's historical data API:
   • Documentation: https://www.bitmex.com/app/apiOverview
   • Or use the bitmex Python package to pull tick-level data for April 17, 2017
   • Save the file as Final_Dataset.csv in the working directory
4. Run analysis.R

---

👤 Author

Swara Dave — MS Financial Engineering, Stevens Institute of Technology