A regression analysis finds a statistically significant, strong relationship between two economic time series that have no plausible underlying connection to each other. This isn't necessarily evidence of a hidden causal link or an interesting coincidence worth investigating — it's frequently a well-understood statistical artifact called spurious regression, which occurs specifically when two independently trending time series are regressed against each other using methods built on assumptions the data doesn't actually satisfy.
Why trending data breaks standard regression assumptions
Standard regression analysis assumes, among other things, that observations are drawn independently, an assumption clearly violated by time series data where each period's value is closely related to the immediately preceding period's value — a property called autocorrelation. When two time series each trend consistently in one direction over an extended period, for entirely separate and unrelated underlying reasons, the shared upward or downward drift alone produces a strong statistical correlation between them, since the regression is effectively picking up on the shared presence of a trend, not on any genuine relationship between what's actually driving each series.
A well-known, deliberately absurd illustration
Researchers studying this phenomenon have compiled examples of genuinely unrelated variable pairs — a country's per-capita cheese consumption and an unrelated statistic like accidental death rates from bedsheet entanglement — that show statistically strong correlations over a shared time period purely because both series happened to trend in the same direction over that period, with no plausible causal connection between them whatsoever. These deliberately absurd examples exist specifically to illustrate how easily spurious correlation can arise between any two trending series, and to caution against treating a strong time-series correlation as meaningful evidence of a real relationship without further testing.
Why this is a genuine, not merely theoretical, risk in applied research
Many economic and business time series genuinely trend over time for structural reasons — population growth, inflation, general technological progress — which means any two such series measured over the same multi-year period carry real risk of showing a spurious correlation purely from shared trend, independent of whether the underlying quantities are meaningfully related. A researcher unaware of this risk can mistake a spurious regression result for a genuine, actionable relationship, particularly when the two variables involved have some superficial plausibility connecting them, making the spurious result feel more credible than a deliberately absurd cheese-consumption example would.
What actually addresses this in practice
Testing whether each time series is stationary (its statistical properties don't change systematically over time) before regressing one against the other is the standard first diagnostic step, since spurious regression specifically arises from non-stationary, trending series. Differencing the data — analyzing period-over-period changes rather than raw levels — often removes the shared trend component that's driving the spurious correlation, revealing whether any genuine relationship remains once the trend itself is accounted for. Cointegration analysis, a more advanced technique, formally tests whether two non-stationary series share a genuine long-run equilibrium relationship, rather than merely sharing an independent trend.
What this means for anyone analyzing time series data
- Test for stationarity before regressing one time series against another, and treat a strong result from non-stationary data with specific skepticism
- Consider analyzing differenced data (period-over-period changes) rather than raw trending levels, to check whether an apparent relationship survives once shared trend is removed
- Be particularly cautious of a strong time-series correlation between two variables that plausibly could each be trending independently for unrelated structural reasons
- Use cointegration analysis rather than simple regression when working with genuinely non-stationary economic or business time series
Spurious regression is a reminder that a statistically significant, high-R-squared result isn't automatically meaningful — with trending time series specifically, it can be close to the expected outcome of the data's shared structure, entirely independent of any real underlying relationship.