Econometrics

Econometrics is the application of statistical and mathematical methods to economic data. It turns economic ideas into numbers: how much households cut spending when interest rates rise, how far demand for petrol falls when its price goes up, whether a training scheme raises earnings, or what GDP growth is likely to be next quarter. Econometricians work in universities, central banks, finance ministries, international bodies such as the IMF, statistics offices, banks and technology firms.

The field sits between economic theory and raw data. Theory supplies the question and a candidate economic model; econometrics supplies the tools to estimate it, test it against evidence and measure the uncertainty in the answer. Its main challenge is that most economic data are observational, not experimental, so separating cause from coincidence requires care. This has driven the growth of methods built around natural experiments and randomized controlled trials.

Suppose a gardener wants to know whether a new fertiliser makes tomatoes grow bigger. She could compare her plants with a neighbour's, but the neighbour's garden may get more sun or water, so the comparison could mislead. She needs a fair way to separate the effect of the fertiliser from everything else going on.

Econometrics is the toolkit for that problem in economic life. It looks at many observations, such as thousands of households or decades of national data, and tries to work out which changes are caused by which factors, and how confident we can be. It also warns when a pattern in the numbers is likely to be a fluke or the result of some hidden influence.

The workhorse method is linear regression, which estimates how one variable moves with others. In its simplest form:

• $y_i$ is the outcome for observation $i$, such as a household's annual spending.
• $x_i$ is an explanatory variable, such as household income.
• $\beta_0$ is the intercept, the predicted outcome when $x_i$ is zero.
• $\beta_1$ is the slope, the change in $y$ for a one-unit change in $x$.
• $\varepsilon_i$ is the error term, capturing everything else that affects $y$.

Ordinary least squares (OLS) chooses $\beta_0$ and $\beta_1$ to minimise the sum of squared errors. For example, if an estimate gives $\beta_0 = 5{,}000$ and $\beta_1 = 0.8$, a household with income of 50,000 is predicted to spend $5{,}000 + 0.8 \times 50{,}000 = 45{,}000$. The slope of 0.8 is an estimate of the marginal propensity to consume. Econometricians also report standard errors and confidence intervals, which often rely on assumptions about the normal distribution of errors.

• Time-series econometrics. Models such as ARIMA and vector autoregressions (VARs) study data over time, for forecasting inflation or measuring the effect of monetary policy shocks. Related tools handle seasonal adjustment and nowcasting with real-time indicators.
• Microeconometrics. Methods for data on individuals and firms, including models of discrete choices such as whether to take a job or buy a product.
• Panel data. Data that follow the same people, firms or countries over time, allowing researchers to control for fixed differences between them.
• Causal inference. Instrumental variables, difference-in-differences and regression discontinuity designs try to recover cause and effect from observational data.
• Financial econometrics. Models of volatility, such as ARCH and GARCH, and of extreme events with fat tails, used in risk measures such as value at risk.

The Norwegian economist Ragnar Frisch is usually credited with coining the word in the 1920s, and he helped found the Econometric Society in 1930, whose journal Econometrica first appeared in 1933. Frisch and the Dutch economist Jan Tinbergen, who built some of the first statistical models of whole economies, shared the first Nobel Memorial Prize in Economic Sciences in 1969. Later prizes recognised Trygve Haavelmo's probability approach (1989), James Heckman and Daniel McFadden's microeconometric methods (2000), Robert Engle and Clive Granger's time-series work (2003), and, in 2021, David Card, Joshua Angrist and Guido Imbens for work on natural experiments and causal inference.

Central banks such as the Federal Reserve use econometric models to forecast and to estimate unobservable quantities like the output gap and the neutral interest rate. Governments use it to evaluate policies; firms use it for pricing and demand forecasting; banks use it in credit risk scoring.

Its limits are widely discussed:

• Correlation is not causation. Omitted variables and reverse causality can make estimates misleading.
• The Lucas critique. Robert Lucas argued in 1976 that relationships estimated under one policy may change when the policy changes, because people adjust their expectations.
• Data quality. Revisions, measurement error and short samples weaken results.
• Specification searching. Trying many models until one "works" can produce false findings, a concern sometimes linked to physics envy in economics.

• The Econometric Society
• Nobel Prize: The Prize in Economic Sciences 1969