Chapter 105: Difference-in-Differences (DiD) for Trading

Overview

Difference-in-Differences (DiD) is a quasi-experimental statistical technique used in econometrics to estimate causal effects by comparing changes in outcomes over time between a treatment group and a control group. Originally developed for policy evaluation, DiD has become a powerful tool in financial markets for measuring the impact of events, regulations, and policies on asset prices.

This chapter implements DiD methods for cryptocurrency trading on Bybit and stock market analysis, demonstrating how causal inference techniques can be applied to identify trading opportunities arising from regulatory changes, policy announcements, and market events.

Theoretical Foundations
DiD Architecture
Mathematical Formulation
Implementation
Trading Strategy
Results and Metrics
References

Theoretical Foundations

What is Difference-in-Differences?

DiD is a research design that estimates causal effects by comparing the difference in outcomes over time between units that receive a treatment and units that do not. The key insight is that by taking a “difference of differences,” we can control for time-invariant unobserved confounders and common time trends.

DiD Estimate = (Y_treated,after - Y_treated,before) - (Y_control,after - Y_control,before)

The Parallel Trends Assumption

The fundamental identifying assumption of DiD is that, in the absence of treatment, the treated and control groups would have followed parallel trends:

E[Y(0)_t | D=1] - E[Y(0)_{t-1} | D=1] = E[Y(0)_t | D=0] - E[Y(0)_{t-1} | D=0]

Where:

Y(0) is the potential outcome without treatment
D indicates treatment status (1 = treated, 0 = control)
The subscripts denote time periods

Why DiD for Trading?

Financial markets offer numerous natural experiments for DiD analysis:

Event Type	Treatment Group	Control Group	Application
Regulatory changes	Affected assets	Unaffected assets	Measure compliance costs
Policy announcements	Target sector	Other sectors	Identify market impact
Exchange listings	Listed token	Similar tokens	Estimate listing premium
Earnings surprises	Announcing company	Industry peers	Isolate company-specific effect
Macroeconomic events	Exposed economies	Unexposed economies	Quantify shock transmission

Advantages Over Simple Event Studies

Traditional event studies measure abnormal returns around an event but struggle with:

Selection bias (which firms are affected is often non-random)
Confounding events occurring simultaneously
Market-wide trends unrelated to the event

DiD addresses these by explicitly constructing a counterfactual using control units.

DiD Architecture

Pre-Period Data                    Post-Period Data
     │                                  │
     ▼                                  ▼
┌─────────────────┐              ┌─────────────────┐
│ Treatment Group │              │ Treatment Group │
│ (Affected)      │              │ (Affected)      │
│    Y_T,pre      │              │    Y_T,post     │
└─────────────────┘              └─────────────────┘
     │                                  │
     │  Δ_T = Y_T,post - Y_T,pre        │
     └──────────────┬───────────────────┘
                    │
                    ▼
            ┌───────────────┐
            │   DiD = Δ_T - Δ_C
            │  (Causal Effect)
            └───────────────┘
                    ▲
     ┌──────────────┴───────────────────┐
     │  Δ_C = Y_C,post - Y_C,pre        │
┌─────────────────┐              ┌─────────────────┐
│  Control Group  │              │  Control Group  │
│  (Unaffected)   │              │  (Unaffected)   │
│    Y_C,pre      │              │    Y_C,post     │
└─────────────────┘              └─────────────────┘

Components

Treatment Identification: Determine which assets are affected by an event
Control Selection: Choose comparable assets unaffected by the event
Pre-Period Analysis: Establish baseline trends and verify parallel trends
Post-Period Analysis: Measure outcomes after treatment
Estimation: Compute DiD estimate with standard errors

Extensions

Staggered DiD: When treatment occurs at different times for different units
Triple Differences (DDD): Adding a third differencing dimension
Synthetic Control DiD: Constructing optimal control weights
Continuous DiD: When treatment intensity varies

Mathematical Formulation

Basic Two-Period Model

For a simple two-period, two-group design:

Y_it = α + β·Treat_i + γ·Post_t + δ·(Treat_i × Post_t) + ε_it

Where:

Y_it: Outcome for unit i at time t
Treat_i: 1 if unit is in treatment group, 0 otherwise
Post_t: 1 if time period is after treatment, 0 otherwise
δ: The DiD estimator (causal effect of treatment)
ε_it: Error term

Generalized Multi-Period Model

With multiple time periods and two-way fixed effects (TWFE):

Y_it = α_i + λ_t + δ·D_it + X_it·β + ε_it

Where:

α_i: Unit fixed effects (time-invariant characteristics)
λ_t: Time fixed effects (common trends)
D_it: Treatment indicator (1 if unit i is treated at time t)
X_it: Time-varying covariates
δ: Average treatment effect on the treated (ATT)

Parallel Trends Test

To validate the parallel trends assumption, estimate:

Y_it = α_i + λ_t + Σ_k δ_k·1(t=k)·Treat_i + ε_it

Where k indexes event time relative to treatment. The pre-treatment coefficients δ_k for k < 0 should be statistically indistinguishable from zero.

Standard Errors

For panel data with potential serial correlation, cluster standard errors at the unit level:

V̂(δ) = (X'X)^{-1} (Σ_i X_i' ε̂_i ε̂_i' X_i) (X'X)^{-1}

Implementation

Python

The Python implementation uses PyTorch and includes:

python/model.py: DiD estimation with regression and matching methods
python/data_loader.py: Data loading for stock (yfinance) and crypto (Bybit) markets
python/backtest.py: Backtesting engine for DiD-based trading strategies

from python.model import DifferenceInDifferences

model = DifferenceInDifferences(
    treatment_col='treated',
    time_col='post_treatment',
    outcome_col='return',
    covariates=['volume', 'volatility', 'market_cap'],
    cluster_col='asset_id',
    n_bootstrap=1000
)

# Fit the model
results = model.fit(panel_data)

# Generate trading signals
signals = model.generate_signals(
    threshold=0.02,      # Minimum effect size
    confidence=0.95,     # Confidence level
    holding_period=5     # Days to hold position
)

Rust

The Rust implementation provides a production-ready version:

src/model/: DiD estimation with efficient matrix operations
src/data/: Bybit API client and feature engineering
src/trading/: Signal generation and strategy execution
src/backtest/: Performance evaluation engine

# Run basic example
cargo run --example basic_did

# Run trading strategy
cargo run --example trading_strategy

# Run event study analysis
cargo run --example event_study

Trading Strategy

Signal Generation

The DiD-based trading strategy identifies opportunities from:

Regulatory Events: New regulations affecting specific sectors
Exchange Listings: Crypto tokens added to major exchanges
Policy Changes: Central bank decisions, fiscal announcements
Corporate Events: Earnings, M&A, management changes

Entry Rules

Long: DiD estimate > threshold AND statistically significant (p < 0.05)
Short: DiD estimate < -threshold AND statistically significant
Flat: Effect not significant or below threshold

Position Sizing

Kelly-criterion inspired sizing based on effect magnitude and confidence:

position_size = (DiD_estimate / std_error) × (confidence - 0.5) × scale_factor

Risk Management

Maximum position size: 15% of portfolio per event
Stop-loss: 2× estimated volatility from control group
Take-profit: 3× DiD effect estimate
Maximum concurrent events: 5
Minimum pre-treatment window: 30 days

Implementation Details

Event Detection: Monitor news feeds, regulatory filings, exchange announcements
Control Construction: Use propensity score matching or synthetic control
Pre-Trend Validation: Verify parallel trends before trading
Continuous Monitoring: Update estimates as new data arrives

Results and Metrics

Evaluation Metrics

Metric	Description
RMSE / MAE	Prediction accuracy for effect magnitude
Coverage Rate	Confidence interval coverage
Sharpe Ratio	Risk-adjusted return
Sortino Ratio	Downside risk-adjusted return
Maximum Drawdown	Worst peak-to-trough decline
Hit Rate	Percentage of correct directional predictions
Profit Factor	Gross profit / Gross loss

Comparison with Baselines

The DiD approach is compared against:

Simple event study (abnormal returns)
Propensity score matching without DiD
Buy-and-hold benchmark
Market-neutral strategy

Empirical Applications

Example use cases included in this chapter:

Crypto Exchange Listings: Measuring the price impact when tokens are listed on Binance/Bybit
Stock Regulation: Impact of SEC enforcement actions on related stocks
Monetary Policy: Effect of FOMC announcements on sector ETFs
ESG Events: Market reaction to sustainability ratings changes

Project Structure

105_diff_in_diff_trading/
├── README.md                  # This file
├── README.ru.md               # Russian translation
├── readme.simple.md           # Simplified explanation (English)
├── readme.simple.ru.md        # Simplified explanation (Russian)
├── Cargo.toml                 # Rust project configuration
├── python/
│   ├── __init__.py
│   ├── model.py               # DiD estimation model
│   ├── data_loader.py         # Stock & crypto data loading
│   ├── backtest.py            # Backtesting framework
│   └── requirements.txt       # Python dependencies
├── src/
│   ├── lib.rs                 # Rust library root
│   ├── model/
│   │   ├── mod.rs             # Model module
│   │   └── did.rs             # DiD implementation
│   ├── data/
│   │   ├── mod.rs             # Data module
│   │   ├── bybit.rs           # Bybit API client
│   │   └── features.rs        # Feature engineering
│   ├── trading/
│   │   ├── mod.rs             # Trading module
│   │   ├── signals.rs         # Signal generation
│   │   └── strategy.rs        # Trading strategy
│   └── backtest/
│       ├── mod.rs             # Backtest module
│       └── engine.rs          # Backtesting engine
└── examples/
    ├── basic_did.rs           # Basic DiD example
    ├── event_study.rs         # Event study analysis
    └── trading_strategy.rs    # Full trading strategy

References

Goodman-Bacon, A. (2021). Difference-in-Differences with Variation in Treatment Timing. Journal of Econometrics, 225(2), 254-277. https://www.sciencedirect.com/science/article/pii/S0304407621001445
Roth, J., Sant’Anna, P. H. C., Bilinski, A., & Poe, J. (2023). What’s Trending in Difference-in-Differences? A Synthesis of the Recent Econometrics Literature. Journal of Econometrics, 235(2), 2218-2244. https://arxiv.org/abs/2201.01194
de Chaisemartin, C., & D’Haultfœuille, X. (2020). Two-Way Fixed Effects Estimators with Heterogeneous Treatment Effects. American Economic Review, 110(9), 2964-2996.
Callaway, B., & Sant’Anna, P. H. C. (2021). Difference-in-Differences with Multiple Time Periods. Journal of Econometrics, 225(2), 200-230.
Abadie, A. (2005). Semiparametric Difference-in-Differences Estimators. Review of Economic Studies, 72(1), 1-19.
Baker, A. C., Larcker, D. F., & Wang, C. C. (2022). How Much Should We Trust Staggered Difference-In-Differences Estimates? Journal of Financial Economics, 144(2), 370-395.
Sun, L., & Abraham, S. (2021). Estimating Dynamic Treatment Effects in Event Studies with Heterogeneous Treatment Effects. Journal of Econometrics, 225(2), 175-199.
MacKinlay, A. C. (1997). Event Studies in Economics and Finance. Journal of Economic Literature, 35(1), 13-39.
De Prado, M. L. (2018). Advances in Financial Machine Learning. Wiley.

License

MIT

Chapter 105: Difference-in-Differences (DiD) for Trading

Chapter 105: Difference-in-Differences (DiD) for Trading

Overview

Table of Contents

Theoretical Foundations

What is Difference-in-Differences?

The Parallel Trends Assumption

Why DiD for Trading?

Advantages Over Simple Event Studies

DiD Architecture

Components

Extensions

Mathematical Formulation

Basic Two-Period Model

Generalized Multi-Period Model

Parallel Trends Test

Standard Errors

Implementation

Python

Rust

Trading Strategy

Signal Generation

Entry Rules

Position Sizing

Risk Management

Implementation Details

Results and Metrics

Evaluation Metrics

Comparison with Baselines

Empirical Applications

Project Structure

References

License