Engle & Granger (2003): Taming Time Series — Volatility and Cointegration

The Challenge of Economic Data Through Time

Economic data has a peculiar property: it moves through time, and time changes everything. Stock prices wander unpredictably. Inflation rises and falls in waves. GDP grows along a trend but fluctuates around it. Standard statistical methods — designed for stable, well-behaved data — can produce nonsensical results when applied to such time series.

Two econometricians developed the tools to handle these challenges. Robert Engle showed how to model volatility that changes over time — calm periods followed by storms. Clive Granger showed how to distinguish genuine long-run relationships from statistical mirages — and discovered that variables that wander individually can be bound together by hidden equilibrium forces.

Engle: ARCH and the Modeling of Volatility

Anyone who watches financial markets knows that volatility clusters. A big price move — up or down — tends to be followed by more big moves. Calm periods are followed by more calm. The 1987 crash was followed by weeks of wild swings; the mid-2000s saw years of unusual tranquility before the 2008 crisis.

Before Engle, econometric models assumed that the variance of errors was constant over time — homoskedasticity. This assumption was clearly wrong for financial data, but no one had a good alternative.

In 1982, Engle introduced the ARCH model (Autoregressive Conditional Heteroskedasticity): the variance of today’s error depends on the size of yesterday’s errors. Large shocks increase expected future volatility; small shocks decrease it.

The key ideas:

• Conditional variance: The volatility at any point in time is not fixed — it’s conditional on recent history. After a market crash, expected volatility is high; after months of calm, it’s low
• Volatility clustering: ARCH captures the empirical fact that large returns (positive or negative) tend to be followed by large returns, and small returns by small returns
• Time-varying risk: Since risk changes over time, so should risk premiums, portfolio allocations, and option prices. ARCH provided the tool to measure this changing risk

Tim Bollerslev (Engle’s student) generalized the model to GARCH (1986), which became the workhorse of financial econometrics — used daily by banks, hedge funds, and central banks worldwide for:

• Risk management: Value-at-Risk (VaR) calculations require volatility forecasts. GARCH provides them
• Option pricing: The Black-Scholes model assumes constant volatility. GARCH-based models allow volatility to change, producing more accurate option prices
• Asset allocation: Investors need to know not just expected returns but expected risk. GARCH forecasts help optimize portfolios dynamically
• Monetary policy: Central banks monitor financial market volatility as an indicator of economic uncertainty

Granger: Cointegration and Spurious Regression

Clive Granger tackled a different but equally fundamental problem: how do you analyze relationships between economic variables that wander over time?

Many economic time series are nonstationary — they don’t fluctuate around a fixed mean but instead drift upward or downward without bound. GDP grows over time. Prices rise with inflation. Stock prices follow random walks.

The danger: if you regress one nonstationary variable on another, you’ll almost always find a “significant” relationship — even if the two variables are completely unrelated. This is spurious regression, and Granger (with Paul Newbold, 1974) showed it was a pervasive problem in economics. Countless published empirical results were statistical illusions.

But Granger also discovered the solution. Some nonstationary variables, while individually wandering, move together in the long run — they’re cointegrated:

• The drunk and her dog: Imagine a drunk woman walking her dog. Both wander randomly — the woman staggers, the dog runs back and forth. But the leash keeps them from drifting too far apart. Each path is nonstationary (unpredictable), but the distance between them is stationary (bounded). The woman and dog are cointegrated
• Economic examples: Consumption and income both grow over time (nonstationary), but their ratio is roughly stable (cointegrated). Short-term and long-term interest rates both wander, but the spread between them is bounded. Exchange rates and relative price levels drift, but purchasing power parity pulls them together in the long run

The Error Correction Model

Granger showed that cointegration implies an error correction mechanism: when cointegrated variables drift apart from their long-run equilibrium, forces pull them back together.

The error correction model captures both:

• Short-run dynamics: How variables respond to shocks in the short run — which may push them away from equilibrium
• Long-run equilibrium: The cointegrating relationship that acts as a gravitational force, pulling variables back toward their equilibrium path

This framework resolved a long-standing tension in economics between short-run Keynesian dynamics and long-run classical equilibrium — they’re not contradictory, they’re complementary aspects of the same system.

Granger Causality

Granger also contributed the concept of Granger causality — a practical test for whether one time series helps predict another. Variable X “Granger-causes” variable Y if past values of X improve forecasts of Y beyond what past values of Y alone provide.

This isn’t true causality in the philosophical sense — but it’s an operationally useful concept that’s been applied thousands of times in economics, finance, neuroscience, and climate science.

Their 2003 Nobel Prize was awarded to Engle “for methods of analyzing economic time series with time-varying volatility (ARCH)” and to Granger “for methods of analyzing economic time series with common trends (cointegration).”

Explain It to a Child

Imagine the weather. Some weeks are stormy — one thunderstorm after another. Other weeks are calm and sunny. Engle figured out how to predict this pattern for stock markets: after a big price swing, expect more big swings. After calm days, expect more calm. His formula helps banks prepare for storms. Now Granger solved a different puzzle. Imagine two friends who both wander around a park randomly. They seem to go wherever they want. But they’re connected by a long rubber band — so they never get too far apart. Even though each friend’s path looks random, the distance between them stays roughly the same. Granger discovered that many economic numbers work this way — they wander on their own but are secretly connected, always pulled back toward each other.