International Association for Quantitative Finance (IAQF) Annual Student Competition
Bitcoin trades simultaneously on dozens of exchanges, quoted in different currencies. We study a subtle but consequential question: does it matter whether you buy BTC with real US dollars or with USDT, a dollar-pegged stablecoin? The short answer is yes — and the gap widens sharply in times of stress.
📄 Full paper: report.pdf
📓 Analysis notebook: notebooks/Q1_cross_currency_basis.ipynb
The IAQF Annual Academic Competition challenges student teams to tackle live problems in quantitative finance. The 2026 edition focuses on cryptocurrency market microstructure and cross-asset dynamics across regulated and unregulated venues.
Our question (Q1): Measure and explain the cross-currency basis — the price difference between BTC quoted in USDT on Binance (the world's largest crypto exchange, offshore) and BTC quoted in USD on Coinbase (the largest US-regulated exchange) — over a 21-day window that includes the March 2023 banking and stablecoin crisis.
When Silicon Valley Bank failed on March 10, 2023, it triggered a brief de-pegging of USDC (a major stablecoin). Even though Tether (USDT) held its peg, all crypto-dollars traded at a discount to real bank dollars. This shows up clearly in BTC prices:
- Normal times: BTC costs ~0.24% more on Coinbase (USD) than Binance (USDT) — a small but persistent premium for regulated, bank-settled dollars
- March 10–13 stress: That gap widened to ~1.5% and stayed wide for hours
Why didn't arbitrageurs close it instantly? Because moving capital between a US-regulated venue and an offshore exchange is slow, costly, and especially risky during a crisis. The gap represents real, measurable funding risk built into the USDT/USD spread.
| Metric | Value |
|---|---|
| Mean log basis (USDT − USD) | −0.24% (USDT persistently cheaper) |
| Peak basis during March 10–13 | ~−1.5% |
| AR(1) half-life of the basis | ~7.5 hours — dislocations are slow to mean-revert |
| Minutes exceeding 0.5% cost band | 13.3% — and 100% one-sided (always USDT discount) |
| Price discovery leader | Binance (60% Hasbrouck information share) |
The one-sidedness is striking: BTCUSDT is never rich enough relative to BTCUSD to justify a trade in the other direction. Arbitrage is constrained, directional, and regime-dependent.
 |
 |
 |
 |
 |
 |
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook notebooks/Q1_cross_currency_basis.ipynbTo re-download Coinbase data:
python -m src.download_coinbase_candles├── report.pdf # Full competition paper
├── notebooks/
│ └── Q1_cross_currency_basis.ipynb # All analysis, tables, and plots
├── src/
│ └── download_coinbase_candles.py # Coinbase REST API downloader
├── data/
│ ├── raw/ # 1-min OHLCV candles (Binance & Coinbase)
│ └── processed/ # Cleaned and aligned datasets
├── figures/ # All output charts (auto-generated by notebook)
├── tests/ # Data integrity tests
└── requirements.txt
- Data: 1-minute OHLCV candles for BTCUSDT (Binance) and BTCUSD (Coinbase), March 1–21, 2023 — ~29,950 aligned observations after cleaning.
-
Basis: Log price difference
$b_t = \log P^{USDT}_t - \log P^{USD}_t$ . Negative = USDT cheaper. - Transaction-cost band: Symmetric round-trip fee of ±0.5% (Binance 0.10% + Coinbase 0.40%).
- Persistence: AR(1) model with regime-specific half-lives; ADF/KPSS stationarity tests.
- Drivers: OLS with HAC (Newey–West) standard errors — stress dummy, realized volatility, volume, lagged basis.
- Cointegration & price discovery: Johansen trace test, VECM, Hasbrouck (1995) information shares, Granger causality.
IAQF Annual Academic Student Competition 2026.