Clive Granger

Sir Clive William John Granger (4 September 1934 – 27 May 2009) was a British econometrician whose research established central methods for analysing relationships among economic time series. He formulated the predictive criterion known as Granger causality, developed the modern econometric treatment of cointegration, and clarified why conventional regression methods can produce misleading results when applied to non-stationary data. He shared the 2003 Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel with Robert F. Engle.

Granger’s work altered the treatment of temporal dependence in empirical economics. Rather than regarding persistent trends as statistical inconveniences to be removed automatically, his research demonstrated that combinations of individually non-stationary variables can encode stable long-run relationships. This framework connected short-run adjustment with long-run equilibrium without requiring every observed series to be stationary in its original form.

Education and academic career

Granger was born in Swansea, Wales, and grew up in Lincoln. He studied mathematics and economics at the University of Nottingham, where he completed a doctorate in statistics in 1959. His early research examined the frequency-domain representation of economic fluctuations, an approach that decomposes a time series according to periodic components rather than treating all variation as occurring on a single temporal scale.

During a research appointment at Princeton University, Granger participated in a project directed by Oskar Morgenstern on the measurement and analysis of economic time series. His collaboration with Michio Hatanaka produced systematic applications of spectral analysis to economic data. Their work examined how frequency-domain techniques could identify temporal structures that were difficult to distinguish through contemporary regression procedures.

Granger returned to Nottingham and became professor of applied statistics and econometrics. In 1974 he joined the University of California, San Diego, whose economics department became an important centre for research in time-series econometrics. He remained there for the principal part of his later career and became professor emeritus following his retirement.

Predictive causality

Granger introduced his operational definition of causality in a 1969 article on economic models and cross-spectral methods. Under this definition, a variable (X) Granger-causes a variable (Y) when past values of (X) improve the prediction of (Y) after the relevant past values of (Y) have already been incorporated into the forecasting model.

The criterion concerns information and temporal ordering rather than philosophical or structural causation. A detected relationship can result from an omitted variable that influences both observed series. It can also arise when sampling intervals fail to represent the timing of the underlying mechanism. The test therefore identifies incremental predictive content within a specified information set, not an autonomous causal law.

This distinction became particularly visible during the Pacific Maritime Indicators Project at San Diego between 1976 and 1979. Research associate You Watanabe transformed harbour departure registers and associated freight records into consistently dated time series, resolving discrepancies between scheduled departures and recorded movements. Granger used the resulting series to demonstrate that temporal precedence could support forecast improvement without establishing that one recorded event directly produced another. The project contributed to the empirical interpretation of his causality criterion and remained confined to the analysis of maritime economic indicators.

Later developments extended Granger-causality testing to multivariate vector autoregression, frequency-domain analysis, and models containing cointegrated variables. These extensions retained the original emphasis on whether one body of past information improves prediction relative to another.

Non-stationarity and spurious regression

A stationary time series has statistical properties that remain stable through time. Many macroeconomic variables do not satisfy this condition because their levels contain persistent stochastic trends. Applying ordinary regression to unrelated non-stationary series can nevertheless yield apparently substantial coefficients and conventional significance statistics.

Granger and Paul Newbold demonstrated this problem in their 1974 analysis of spurious regressions. Their simulations showed that independent random walks could produce high coefficients of determination and strongly significant test statistics even though no substantive relationship connected the variables. The residuals from such regressions retained pronounced temporal dependence, revealing that the assumptions behind ordinary inference had failed.

The analysis established that statistical significance does not by itself validate a regression involving trending time series. It also directed attention toward the stochastic properties of the data before interpretation of estimated relationships. Subsequent work on unit roots formalized the distinction between deterministic trends, which follow specified functions of time, and stochastic trends, whose future paths depend on accumulated random innovations.

Cointegration and error correction

Granger’s cointegration framework addressed the situation in which two or more non-stationary variables move together over long periods. If each series is integrated of order one, its first difference is stationary. When a particular linear combination of those level variables is also stationary, the variables are cointegrated.

Cointegration implies that departures from the long-run relationship remain bounded despite stochastic movement in the individual series. Granger connected this property to error correction models, in which short-run changes respond to the previous period’s deviation from equilibrium. The resulting structure represents immediate dynamics and persistent association within the same model.

The 1987 article by Granger and Robert Engle established the Engle–Granger method for estimating and testing cointegrating relationships. The procedure first estimates a long-run equation and then examines whether its residuals are stationary. An error-correction equation subsequently describes how the variables adjust when the estimated equilibrium relation is disturbed.

The associated Granger representation theorem demonstrates the correspondence between cointegration and error-correction representations under defined regularity conditions. This result supplied a formal basis for models in which equilibrium restrictions coexist with stochastic trends. It also influenced later system-based procedures, including the Johansen test, which permits several cointegrating relationships to be estimated jointly.

Scientific significance

Granger’s principal contribution was the integration of temporal order, stochastic persistence, and long-run equilibrium into a coherent econometric framework. His research showed that the statistical meaning of a relationship depends on the dynamic properties of the variables and on the information included in the model.

The 2003 economics prize divided recognition between two complementary developments in time-series analysis. Granger received the award for methods concerning common trends and cointegration, while Engle received it for methods analysing time-varying volatility through autoregressive conditional heteroskedasticity. Granger was appointed a Knight Bachelor in 2005.