The Economic Impact of Low- and High-Frequency Temperature Changes

This paper demonstrates that low-frequency temperature variations, distinct from high-frequency fluctuations, are significantly associated with reduced economic growth—particularly in European and international contexts—where the post-1980 rise in this component is estimated to have lowered growth by approximately 1.3 percentage points.

The Gist

Imagine the Earth's temperature isn't just a single, steady number that goes up and down. Instead, think of it like a symphony playing in the background of our economy. This symphony has two very different types of music happening at the same time:

1. The Slow, Deep Bass (Low-Frequency): This is the long-term trend. It's the slow, creeping rise in temperature we call climate change. It's like the bass line of a song; you might not notice it changing every second, but over years and decades, it sets the entire mood.
2. The Fast, Jittery Drums (High-Frequency): This is the daily weather. It's the heatwaves, the cold snaps, the El Niño storms, and the sunny days. It's the fast, erratic drumming that makes you jump or sweat right now, but it usually settles down quickly.

The authors of this paper asked a simple but crucial question: Do these two types of "music" hurt our economy in different ways?

The Big Discovery: It Depends on Where You Live

The researchers looked at data from three different "concert halls": the United States, Europe, and the rest of the world. They found that the "bass" and the "drums" affect these places differently.

1. The United States: The Drums Matter More

In the U.S., the economy seems to be mostly bothered by the fast drums (extreme weather events like hurricanes or heatwaves).

• The Analogy: Imagine a runner trying to sprint while someone keeps tripping them with a stick every few minutes. The runner (the economy) gets frustrated and slows down right then, but once the stick is gone, they can keep going.
• The Finding: The slow, long-term warming (the bass) hasn't actually slowed down U.S. economic growth in a statistically significant way yet. The U.S. economy is resilient to the slow creep of heat, but it stumbles when the weather gets crazy and extreme.

2. Europe and the World: The Bass is Crushing the Band

In Europe and the international panel (50 countries), the story is different. Here, the slow bass (the long-term warming trend) is the real problem.

• The Analogy: Imagine a runner trying to run a marathon, but the ground beneath them is slowly turning into thick, sticky mud. They aren't being tripped by sticks; they are just getting heavier and slower with every step because the environment itself is changing.
• The Finding: As the long-term temperature trend rises, economic growth is dropping significantly. The paper estimates that the warming trend since 1980 has shaved about 1.3 percentage points off annual economic growth in these regions. It's a slow, persistent drain on wealth that doesn't just "bounce back" like a heatwave does.

Why Did It Take So Long to Figure This Out?

The paper also highlights a major headache in how economists usually do their math.

The "One-Way" Mistake:
Most previous studies looked at data like a single file of people standing in a line. They assumed that if one person had a bad year, it didn't affect the person next to them.

• The Reality: In the real world, everyone is in the same room. If it rains on the whole country, everyone gets wet at the same time.
• The Fix: The authors realized that standard math tools were "blind" to these shared problems. They developed a new, more sophisticated way of looking at the data (like using a 3D scanner instead of a 2D photo) that accounts for the fact that all states or countries are reacting to the same global weather patterns simultaneously.

The Takeaway for Us

This research tells us that we can't treat "weather" and "climate" as the same thing when it comes to money.

• Short-term weather (High-Frequency) causes immediate, sharp shocks. We need to build better emergency responses for this (like storm shelters or heat relief).
• Long-term climate (Low-Frequency) is a slow, grinding erosion of our potential. It's not just about surviving a storm; it's about the fact that the soil is slowly becoming less fertile for the next 50 years.

The Bottom Line:
If you live in the U.S., your wallet is currently most at risk from the storms. But if you look at the rest of the world (and likely the U.S. in the future), the slow, steady rise in temperature is quietly stealing a massive chunk of our future economic growth. We need to stop just looking at the daily weather forecast and start paying attention to the long-term trend, because that's where the real economic damage is hiding.

Technical Summary

1. Problem Statement and Motivation

The paper addresses a critical gap in the literature regarding the economic impact of climate change. Existing studies often treat temperature variations as a single aggregate or focus primarily on short-term weather shocks (high-frequency) or long-term climate trends (low-frequency) in isolation. The authors argue that these two components have distinct dynamic properties and may transmit differently to economic activity:

• Low-Frequency ($L^0_X$): Represents slowly evolving trends (akin to climate), generating persistent, long-lasting effects.
• High-Frequency ($H^0_X$): Represents volatile deviations (akin to weather, e.g., heatwaves, El Niño), generating short-lived effects.

The central research question is whether these distinct components have heterogeneous impacts on economic growth ($\Delta Y$) and how to properly model them given the challenges of unobserved heterogeneity and common time effects in panel data.

2. Methodology

A. Data Decomposition and Estimation

The authors decompose observed temperature $X$ into unobserved low- and high-frequency components ($X = L^0_X + H^0_X$).

• Parametric Approach: They utilize an Unobserved Components (UC) model with long memory. $L^0_X$ is modeled as a fractionally integrated process $I(d)$, while $H^0_X$ is a stationary process (e.g., AR(1)). Parameters are estimated via Maximum Likelihood.
• Non-Parametric Approach: To avoid reliance on specific parametric assumptions, they employ the Mueller-Watson (MW) procedure, which projects the series onto a set of periodic functions to isolate low-frequency trends. They also compare this with Hodrick-Prescott (HP) and Hamilton (JH) filters.
• Validation: Monte Carlo simulations calibrated to US temperature data confirm that the MW procedure (specifically with $q=8$) provides robust estimates of the low-frequency trend with minimal approximation error.

B. Econometric Framework

The authors analyze three panels:

1. U.S. Panel: 48 contiguous states (1964–2023).
2. European Panel: 20 countries (subset of the international panel).
3. International Panel: 50 relatively developed countries (1953–2018).

They estimate the impact on real per capita GDP/GSP growth ($\Delta Y$) using two main regression specifications:

1. Static Model: $\Delta Y_{it} = m_{it} + \beta_L \hat{L}_{X,it} + u_{it}$
2. Dynamic Model (ARDL): Includes lags of $\Delta Y$ and high-frequency components ($\hat{H}_{X,it}$).

Handling Unobserved Heterogeneity:
A key methodological contribution is the treatment of unobserved common factors. The authors compare three specifications:

• FE (Fixed Effects): Unit fixed effects only.
• AFE (Additive Fixed Effects): Unit + Time fixed effects.
• IFE (Interactive Fixed Effects): $\gamma_i + \lambda_i F_t$, where $F_t$ is a common factor and $\lambda_i$ are factor loadings. This allows for heterogeneous responses to common time shocks.

Inference and Standard Errors:
The paper highlights a critical flaw in standard panel data inference for this context. The authors demonstrate via simulation that one-way clustered standard errors (clustering only by unit) lead to severe size distortions (over-rejection of the null) when common time effects are present. They advocate for:

• Two-way clustering (unit and time) for FE/AFE models.
• Bootstrap inference (fixed-design, percentile wild bootstrap) for IFE models to account for the uncertainty in estimating latent factors.

3. Key Results

A. Characterization of Temperature Components

• Persistence: The low-frequency component ($\hat{L}_X$) is highly persistent (non-stationary) with a periodicity of roughly 32 years in the US data.
• Spatial Heterogeneity: There is significant heterogeneity in the volatility of components. Coastal US states show stronger low-frequency trends, while interior states exhibit higher high-frequency volatility.
• Common Factors: A common low-frequency factor exists across all panels, strongly correlated with the aggregate temperature trend, which has risen significantly since the late 1970s.

B. Economic Impact Estimates

The results vary significantly by region and model specification:

1. United States Panel:

• Low-Frequency Effect: Under simple Fixed Effects (FE) models, a negative effect is found, but it becomes statistically insignificant when using two-way clustered standard errors or Interactive Fixed Effects (IFE).
• High-Frequency Effect: The high-frequency component shows a marginally significant negative effect on growth in the US, suggesting US economic growth is more sensitive to short-term weather shocks (heatwaves, disasters) than long-term trends.
• Conclusion: The long-run impact of the rising low-frequency temperature component on US growth is not statistically different from zero.

2. European and International Panels:

• Low-Frequency Effect: In contrast to the US, the low-frequency component has a large, statistically significant negative effect on GDP growth in both European and International panels.
  • Magnitude: In the European panel (IFE specification), the coefficient is approximately -0.9. Given a 1.48°C rise in the low-frequency component (1980–2018), this translates to a reduction in annual GDP growth of roughly 0.035 percentage points, accumulating to a total reduction of ~1.3 percentage points over the period.
  • International Panel: The coefficient is even larger (approx. -1.28), implying a 1°C rise in the low-frequency component reduces growth by >1.28 percentage points.
• High-Frequency Effect: The first-order effect of high-frequency changes is generally insignificant in these panels, though there is evidence of nonlinearity (interaction between high- and low-frequency components).

C. Dynamic Responses

Using Impulse Response Functions (IRFs) based on the UC model:

• Shocks to the high-frequency component dissipate within 3–5 years.
• Shocks to the low-frequency component have persistent effects on output growth, lasting 20+ years. In countries with high long-memory parameters ($d > 1$), these effects are amplified.

4. Key Contributions

1. Frequency Decomposition: The paper rigorously separates temperature into low- and high-frequency components, demonstrating that they have distinct and regionally heterogeneous economic impacts.
2. Methodological Advancement in Inference: The authors provide strong evidence that standard one-way clustering in panel regressions of climate data leads to spurious significance. They establish Interactive Fixed Effects (IFE) combined with bootstrap inference as the preferred methodology for handling common time effects and factor structures in climate-economy panels.
3. Regional Heterogeneity: The findings challenge the "one-size-fits-all" view of climate impacts. They suggest that while the US economy may be adapting to or insulated from long-term temperature trends (perhaps due to infrastructure or economic structure), European and International economies suffer significant long-run growth penalties from the same trends.
4. Nonlinearity: The paper identifies that the impact of high-frequency weather shocks may depend on the level of the low-frequency temperature trend, introducing a new form of nonlinearity in climate-economy modeling.

5. Significance and Policy Implications

• Mitigation Policy: The findings suggest that the economic costs of global warming are not uniform. The significant negative impact on international and European growth implies that the "slow burn" of rising baseline temperatures (low-frequency) poses a severe threat to long-term economic potential, distinct from the immediate damage of extreme weather events.
• Adaptation: The lack of a significant long-run effect in the US panel suggests that adaptation mechanisms (e.g., infrastructure, technology, economic diversification) may be more effective in the US than in other regions, or that the US is less sensitive to the specific nature of the temperature trend observed.
• Modeling Best Practices: The paper serves as a cautionary guide for econometricians studying climate economics, emphasizing the necessity of controlling for common time factors and using robust inference methods (two-way clustering or bootstrapping) to avoid overestimating the statistical significance of climate variables.

In summary, the paper concludes that low-frequency temperature increases are associated with a significant reduction in economic growth globally (approx. 1.3 percentage points over the post-1980 period), but this effect is masked in the US data by high-frequency volatility and requires sophisticated econometric techniques to uncover.