What Is Econometrics? Models and Methods Explained

Search Fintrev

Search our guides, reviews and glossary.

Econometrics is the branch of economics that applies statistical and mathematical methods to real-world data in order to measure economic relationships, test economic theories and forecast future outcomes. Economic theory might say that households spend more when their income rises. Econometrics asks how much more, whether the data actually support the claim, and how confident anyone can be in the answer.

What Is Econometrics?

The field is usually split into two parts. Theoretical econometrics develops and studies the methods themselves: which estimators work, under what assumptions, and how they behave when those assumptions fail. Applied econometrics uses those methods on actual data, either to test an existing theory or to work from the data toward a new hypothesis.

The discipline took shape in the early 20th century. Norwegian economist Ragnar Frisch is credited with coining the word, the Econometric Society was founded in 1930, and Frisch shared the first Nobel Memorial Prize in Economic Sciences in 1969.

Econometrics vs. Mathematical Economics and Statistics

Mathematical economics expresses theory as exact functions. A consumption function might be written C = a + bY, meaning consumption is fully determined by income. An econometric function adds an error term: C = a + bY + ε. The error term (ε) captures everything the model leaves out, such as tastes, wealth, measurement mistakes and plain randomness. That one addition is what makes the relationship something to estimate rather than assume.

Econometrics also differs from general statistics. It borrows statistical tools, but its data are mostly observational rather than experimental. Economists rarely get to run controlled trials on interest rates or tax policy, so much of the field is about extracting credible estimates from data nobody designed for the purpose.

How Econometrics Works

The classical methodology of econometrics follows a sequence of steps:

1. State the theory or hypothesis. For example, higher unemployment leads to lower stock prices.
2. Specify a mathematical model: the variables involved and the form of the relationship.
3. Specify the econometric model by adding the error term and the assumptions about it.
4. Collect data, such as historical prices for an index like the S&P 500 and unemployment rates.
5. Estimate the parameters, usually with regression.
6. Test the hypothesis for statistical significance and consistency with theory.
7. Forecast or evaluate policy with the fitted model.

In the unemployment example, the stock market return is the dependent variable and the unemployment rate is the independent (explanatory) variable. A simple linear regression model looks like this:

Y = β₀ + β₁X + ε

Here β₀ is the intercept, β₁ is the slope (the estimated change in Y for a one-unit change in X), and ε is the error term. OLS chooses β₀ and β₁ to minimize the sum of squared residuals, the squared vertical distances between each data point and the fitted line. Adding explanatory variables, such as GDP growth and inflation alongside unemployment, gives multiple linear regression, the workhorse of applied econometrics.

Types of Data and Models

Econometric data come in three main forms. Cross-sectional data cover many units at one point in time (thousands of households in a single year). Time-series data follow one unit over time (monthly inflation rates by year). Panel data combine both, tracking many units across many periods.

The model depends on the data and the question:

• OLS and multiple regression for continuous outcomes.
• Logit and probit models for yes-or-no outcomes, such as whether a borrower defaults or a worker loses a job.
• Time-series models such as autoregressive models for forecasting a variable from its own past.
• Simultaneous equations models for systems where variables determine each other, such as supply and demand.

Interpreting Econometric Results

A regression output contains a few numbers that matter most:

• Coefficient: the estimated size and direction of the effect.
• Standard error: how precisely the coefficient is estimated.
• t-statistic: the coefficient divided by its standard error, used to test whether the true value could be zero (the null hypothesis).
• p-value: the probability of seeing an estimate at least this large if the null hypothesis were true. A p-value below a chosen threshold, often 0.05, is called statistically significant.
• R-squared: the share of variation in the dependent variable explained by the model, from 0 to 1.

A high R-squared does not mean the model is correct, and a significant coefficient does not mean the effect is large enough to matter in practice. Statistical significance and economic significance are separate questions.

Example of Econometrics

Hypothetical example. An economist wants to measure how household spending responds to income. Using survey data on 2,000 households, she estimates:

Consumption = 5,000 + 0.75 × Income

The slope of 0.75 is the estimated marginal propensity to consume: each additional $1,000 of annual income is associated with about $750 of additional spending. With a standard error of 0.05 on the slope, the t-statistic is 0.75 ÷ 0.05 = 15. That is far above the roughly 1.96 needed for significance at the 5% level. An R-squared of 0.60 means income accounts for 60% of the variation in spending across these households. The remaining 40% comes from factors outside the model.

A policymaker could use an estimate like this to project how much a tax rebate might feed into spending.

Estimators, Autocorrelation and Endogeneity

Estimators

An estimator is a rule for calculating an unknown population value from a sample. Official unemployment figures, for example, are estimated from random samples of households. OLS is itself an estimator. Good estimators are unbiased (correct on average), consistent (closer to the truth as the sample grows) and efficient (lowest variance among comparable estimators).

Autocorrelation

Autocorrelation, also called serial or lagged correlation, is correlation between a variable and its own past values. It is common in time-series data: this quarter’s GDP growth tends to resemble last quarter’s. In regression, autocorrelated error terms do not bias OLS coefficients by themselves, but they make the standard errors wrong, which can make results look more significant than they are.

Endogeneity

A variable is endogenous when it is correlated with the error term, often because it is determined within the system being studied. For example, price and quantity are set together in a market. Omitted variables and causation running in both directions can also cause it. Endogeneity biases OLS estimates, and econometricians address it with methods such as instrumental variables, which use an outside variable that affects the explanatory variable but not the outcome directly.

Limitations of Econometrics

Correlation is not causation. Drowning deaths in swimming pools rise as GDP grows. A booming economy does not make people drown. More households buy pools when incomes are rising. Regression detects association, and a causal claim needs a credible mechanism and a research design that rules out alternatives.

Theory still matters. John Maynard Keynes was among the economists who criticized early econometric work for leaning on statistical correlations rather than economic reasoning. Results that cannot be explained by a theory, existing or new, are fragile.

Data and specification problems. Economic data are revised, measured with error and often short. Leaving out a relevant variable, choosing the wrong functional form or regressing one trending series on another can produce spurious results that look highly significant.

Relationships change. A model estimated on one period may break down when policy, technology or behavior shifts, which limits how far forecasts can be pushed.

FAQs

The Bottom Line

Econometric results matter to anyone who relies on economic estimates, from policymakers projecting a tax rebate's effect to readers weighing claims about markets. A regression can show association, but a causal claim needs sound theory, a credible mechanism and a design that rules out alternatives. Statistical significance and practical importance are separate questions, and relationships estimated on one period can break down when policy or behavior shifts.