## Rising Temperature, Nuanced Effects: Evidence from Seasonal and Sectoral Data

## Source details

**Canonical URL:** [Rising Temperature, Nuanced Effects: Evidence from Seasonal and Sectoral Data](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024202-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024202-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024202-print-pdf.pdf.json)

---

### Abstract and central findings
- Dataset and approach:
  - Uses quarterly temperature and sectoral value-added data for a large sample of advanced economies (AEs) and emerging markets and developing economies (EMDEs).
- EMDEs:
  - Hotter spring and summer temperatures reduce growth in real value-added of manufacturing, and most significantly, of agriculture.
  - A warmer winter boosts agricultural value-added.
  - A 1-Celsius degree hotter spring reduces year-on-year (yoy) growth in value-added of agriculture by about 0.8 percentage points in the same spring and by more than 1 percentage point in the following summer and fall.
  - A hotter summer also reduces growth in agricultural value-added persistently; a hotter fall does not have a significant effect.
- AEs:
  - A hotter spring hurts growth in real value-added of all considered sectors: services, manufacturing, and agriculture.
  - A warmer winter boosts agriculture (persistently), but not manufacturing and services.
  - Because agriculture is small in AEs, a warmer winter has a negligible impact on the aggregate economy.
- Commonalities:
  - The negative effect of a hotter spring is larger and more persistent than the positive effect of a warmer winter.
  - At peak, the magnitude of the negative spring aggregate effect is quite similar for EMDEs and AEs.
  - Agriculture is the most affected sector by temperature.
- Time evolution:
  - Adverse impacts of hotter temperatures in advanced economies have accentuated in recent decades; the worsening impacts for AEs are especially significant, large, and consistent across seasons, and were largely driven by services.
  - Worsening impacts for EMDEs are less significant and mostly driven by higher spring temperature.
- Contextual climate facts:
  - The global average temperature is already about 1.1 degree Celsius higher than the pre-industrial level.
  - Example: Highest daily temperature in Washington D.C. in 2021 ranges from 6 degrees Celsius in the winter to the 35 degrees Celsius in the summer; the average annual temperature for Washington D.C. is about 20 degrees Celsius.

### Sectoral and seasonal magnitudes (sample, measurement, and summary statistics)
- Sample and coverage:
  - Real quarterly value-added dataset: 74 countries (30 are AEs and 44 are EMDEs); time coverage spans the 1990-2019 period (pandemic years dropped).
  - Real quarterly GDP: 87 countries (30 are AEs and 57 are EMDEs); time coverage 1990-2019.
  - ERA5 temperature data collected between 1980 and Q2 of 2020 (ERA5 data available to July 9, 2020).
- Selected summary statistics for YoY growth (%) in real quarterly value-added and GDP, 1990-2019 (after winsorization of top and bottom 1%):
  - Agriculture value-added (Number of countries 74): Min -26.418; p1 -19.654; p25 -2.561; Median 2.138; p75 5.722; p99 27.206; Max 36.047.
  - Manufacturing value-added (74): Min -15.397; p1 -12.134; p25 0.105; Median 2.971; p75 6.261; p99 19.132; Max 22.285.
  - Services value-added (74): Min -6.857; p1 -4.573; p25 1.776; Median 3.443; p75 5.551; p99 14.001; Max 17.768.
  - (Agri+Man+Serv) value-added (74): Min -7.659; p1 -5.126; p25 1.692; Median 3.340; p75 5.349; p99 12.285; Max 14.437.
  - GDP (87): Min -9.014; p1 -6.309; p25 1.623; Median 3.528; p75 5.615; p99 12.334; Max 14.100.
- Seasonal definitions:
  - Quarter 1 (January to March) = winter; Quarter 2 (April to June) = spring; Quarter 3 (July to September) = summer; Quarter 4 (October to December) = fall.
  - Southern hemisphere: winter assigned July to September; spring October to December; summer January to March; fall April to June.
- Temperature construction:
  - Maximum daily temperature from ERA5 averaged across gridcells within a country to construct daily country-level data; daily values averaged across days to generate quarterly seasonal mean of daily maximum temperature; temperature is in Celsius.
- Temperature summary, 1980-Q2 2020 (medians across 219 countries/territories):
  - Temperature Level (Celsius) medians: Winter (Q1) 25.08; Spring (Q2) 27.16; Summer (Q3) 27.80; Fall (Q4) 25.77.
  - YoY Change in Temperature (Celsius) medians: Winter (Q1) 0.028; Spring (Q2) 0.031; Summer (Q3) 0.031; Fall (Q4) 0.031.
- Typical median yearly increases in seasonal temperature: about 0.028 to 0.031 Celsius degrees a year (translating to about 1.2 Celsius degrees over 40 years).
  - Example country slopes: Vietnam slope 0.03; Spain slope 0.072.

### Theoretical motivation and empirical specification
- Two modeled channels (based on Dell et al., 2012):
  - Level effect (equation (1)): Y_{s,q} = e^{β_s T_q} A_{s,q}^{α}; β_s captures season-specific and sector-specific level effects.
  - Growth effect (equation (2)): log(A_{s,q}) − log(A_{s,q−4}) = g + λ_s T_q + σ_s T_{q−4}; λ_s and σ_s capture current-season and same-season-last-year temperature effects on yoy seasonal productivity growth.
  - Combined implication (equation (3)) motivates inclusion of current-season temperature T_q and same-season-last-year temperature T_{q−4} in empirical models.
  - Special case with λ_s = σ_s = 0 reduces to growth as a function of year-on-year change in temperature: log(Y_{s,q}) − log(Y_{s,q−4}) = αg + β (T_q − T_{q−4}).

### Empirical specification and controls (local projections)
- Baseline local-projection specification estimates dynamic impact of quarter q’s temperature on YoY sectoral value-added growth up to h = 1,2,3 quarters ahead.
- Dependent variable: Δ(VA)_{q+h,c,s} = sector s’ yoy quarterly value-added growth of quarter q+h.
- Main regressor: Temp_{q,c} = country c’s average temperature in quarter q. Coefficient of interest β_{s,h}.
- Controls X_{q,c,h} include:
  - Temp_{q+f,c} for f = −4, −3, −2, −1, 1, ..., h.
  - Prec_{q+f,c} for f = −4, −3, −2, −1, 0, 1, .., h.
  - Lagged growth Δ(VA)_{q+f,c,s} for f = −4, −3, −2, −1.
- Fixed effects:
  - Country fixed effects fe_c and year-quarter fixed effects fe_q (1990–2019).
- Regressions run separately for agriculture, manufacturing, and services.
- Note: Short time-series dimension focuses on changes in seasonal temperature relative to country averages.

### Main findings (seasonal and sectoral effects, quantitative highlights)
- General note: Figure 2 reports dynamic impacts of a 1-Celsius degree higher seasonal temperature on YoY growth; results distinguished by AEs and EMDEs.
- Agriculture:
  - EMDEs:
    - A 1-Celsius degree hotter spring reduces YoY growth in agricultural value-added in the same quarter by about 0.8 percentage points.
    - It also reduces YoY growth in agricultural value-added for the following fall and winter by more than 1 percentage point.
    - A hotter summer has a contemporaneous adverse effect, with subsequent fall decline of about 1 percentage points.
    - A 1-Celsius degree warmer winter increases growth in agriculture’s value-added (less statistically significant).
  - AEs:
    - A 1-Celsius degree hotter spring reduces YoY growth in agriculture in the same season and the following one (smaller magnitudes than EMDEs).
    - Hotter summer reduces agricultural growth in summer and following fall.
    - A 1-Celsius degree warmer winter increases agricultural value-added significantly and persistently for three quarters.
- Manufacturing:
  - Hotter springs and summers yield lower YoY growth in manufacturing for both AEs and EMDEs, although summer effects are not statistically significant.
  - Effects smaller than in agriculture and tend to be short-lived.
  - Hotter winters appear to boost manufacturing growth.
- Services:
  - Hotter temperatures have the lowest impact on services.
  - For EMDEs, most seasons show non-significant effects on services.
  - For AEs, a hotter spring hurts YoY growth in services in the spring and subsequent seasons; the impact is small but statistically significant.
- Aggregate value-added and GDP:
  - EMDEs: Hotter spring reduces YoY growth in aggregate value-added contemporaneously and in following season (summer); effect becomes insignificant by fall.
  - AEs: Hotter spring reduces YoY growth in aggregate value-added in the same spring and persistently across three subsequent seasons.
  - Peak magnitude of negative spring aggregate effect is quite similar for EMDEs and AEs.
  - Similar results hold for GDP growth; effects are more persistent for AEs.

### Mechanisms and interpretation
- Agriculture channels:
  - Spring critical for planting and crop growth; high temperatures cause morpho-anatomical, physiological and biochemical plant changes, heat stress, reduced photosynthesis, increased evaporation and soil moisture loss, increased weeds and pests, and reduced pollination.
  - Hot summers coincide with harvesting; extreme heat during harvest can disrupt production.
  - These processes explain protracted growth effects, particularly in EMDEs.
- Manufacturing channels:
  - Direct effects: worsened working conditions (reduced labor productivity, higher absenteeism), especially where temperature control is limited.
  - Indirect effects: linkages to agriculture (inputs tied to agricultural output).
- Services channels:
  - Many service activities occur indoors, reducing sensitivity to outdoor temperature.
  - In AEs, outdoor-oriented services (gardening, landscaping, outdoor recreation, events) may explain small negative spring effects.

### Impacts by climate zones and income (interactions)
- Extended specification includes Temp × Tropics (tropical countries south of Tropic of Cancer and north of Tropic of Capricorn) and Temp × AEs.
- Key results:
  - For aggregate value-added growth, AEs do not appear more resilient to higher temperature than EMDEs after controlling for climate zone: AEs*temperature interactions are either not significantly different from zero or sometimes negative.
  - For agriculture specifically, AEs fare better than EMDEs once geographic location (climate zone) is controlled for.

### Changes over time (decadal analysis)
- Sample split into 1990s (1990-1999), 2000s (2000-2009) and 2010s (2010-2019); decadal dummy interacted with temperature.
- AEs:
  - Adverse impact of a 1-Celsius increase on aggregate value-added growth is more pronounced in the 2010s compared to the 1990s.
  - The interaction term 2010s*temperature is largely negative and significant for all seasonal temperature.
  - Increasing impact largely driven by services.
  - Possible interpretations: effect of a 1-Celsius temperature increase becoming more severe as weather gets hotter; AE services may be shifting toward more outdoor-oriented industries.
- EMDEs:
  - Decadal impacts are more localized to specific seasons.
  - Impact of spring temperature on GDP growth more negative since the 2000s.
  - No statistically significant changes over time for other seasons.
- Quantitative reiteration:
  - For EMDEs, a 1-Celsius degree hotter spring reduces yoy growth in agricultural value-added in the same quarter by about 0.8 percentage points and by more than 1 percentage point for the following fall and winter.

### Robustness checks and extensions
- Robustness 1 (no winsorizing): including top and bottom 1% of growth yields similar patterns with larger magnitudes and larger standard errors.
  - For EMDEs, magnitude of agriculture impact is close to 2% in this exercise.
  - A hotter spring still hurts services and manufacturing in AEs.
- Robustness 2: control for lags of yoy growth in value-added up to four quarters of all sectors; quantitative results unchanged.
- Extreme temperature analysis:
  - Specification includes extreme_heat_q,c dummy = 1 if temperature above 70th percentile of country-quarter historical distribution.
  - Extreme heat yields larger declines in growth relative to intermediate temperature values across sectors.
  - Agriculture: extreme heat in spring, summer and fall leads to more severe growth slowdown.
  - Manufacturing: extreme heat hurts value-added growth in summers.
  - Services: extremely hot summer helps value-added growth in services (possibly indoor activity or beach-going).

### Conclusions and policy implications
- Sectoral and seasonal heterogeneity:
  - EMDEs:
    - Hotter spring and summer temperatures reduce growth in manufacturing and, most significantly, agriculture.
    - A 1-Celsius degree hotter spring reduces yoy agricultural value-added in the same quarter by about 0.8 percentage points and by more than 1 percentage point for the following fall and winter.
    - A warmer winter boosts agricultural activity.
  - AEs:
    - A hotter spring hurts growth in services, manufacturing and agriculture.
  - For both groups:
    - Negative effect of a hotter spring is larger and more persistent than positive effect of a warmer winter.
    - For EMDEs, the spring effect is driven more by agriculture and manufacturing; for AEs, by agriculture, manufacturing and services.
- Broader implications:
  - Timing of climate shocks and sectoral composition matter when gauging climate change impacts.
  - Larger impacts on agriculture—where many low-income workers are employed—point to potentially adverse consequences on poverty and inequality.
  - The impact of temperature on economic activity is mainly driven by extreme heat; above-mean temperatures are particularly negative for economic activity.
  - Adverse impacts of hotter temperatures in AEs and, to a lesser extent, EMDEs have accentuated in recent decades; in AEs this trend is largely driven by services.
- Policy recommendations:
  - Invest in adaptation to improve resilience to changing climate conditions.
  - Adjust economic structure to reduce susceptibility to temperature over time, especially in service-intensive advanced economies.

### Estimation details and selected coefficient estimates (seasonal horizons 0 to +3)
- Estimation conventions:
  - Robust standard errors in parentheses.
  - Significance: * p<0.1, ** p<0.05, *** p<0.01.
- Spring (key coefficients):
  - Temperature: Horizon 0: -0.333 (0.277); Horizon +1: -0.561* (0.308); Horizon +2: 0.0625 (0.300); Horizon +3: 0.880*** (0.308).
  - Temperature*extreme heat dummy: Horizon +2: -0.0442** (0.0190); Horizon +3: -0.0520** (0.0242).
  - Observations: Horizon 0: 1,256; +1: 1,243; +2: 1,248; +3: 1,192.
  - R-squared: Horizon 0: 0.627; +1: 0.504; +2: 0.462; +3: 0.227.
- Fall (key coefficients):
  - Temperature: Horizon 0: 0.112 (0.153); Horizon +1: 0.550** (0.229); Horizon +2: 0.451 (0.282); Horizon +3: 0.782** (0.331).
  - Temperature*extreme heat dummy: Horizon +2: -0.0571* (0.0289).
  - Observations: Horizon 0: 1,197; +1: 1,196; +2: 1,195; +3: 1,178.
  - R-squared: Horizon 0: 0.599; +1: 0.250; +2: 0.243; +3: 0.273.
- Summer (key coefficients):
  - Temperature: Horizon 0: -0.375 (0.282); Horizon +1: -0.818** (0.312); Horizon +2: 0.735 (0.471); Horizon +3: 1.241*** (0.467).
  - Temperature*extreme heat dummy: Horizon +2: -0.0496** (0.0212); Horizon +3: -0.0484** (0.0220).
  - Observations: Horizon 0: 1,254; +1: 1,257; +2: 1,199; +3: 1,197.
  - R-squared: Horizon 0: 0.632; +1: 0.522; +2: 0.239; +3: 0.245.
- Winter (key coefficients):
  - Temperature: Horizon 0: 0.611*** (0.166); Horizon +1: 0.364* (0.193); Horizon +2: 0.715*** (0.207); Horizon +3: 0.0360 (0.193).
  - Temperature*extreme heat dummy: no significant positive interaction reported.
  - Observations: Horizon 0: 1,263; +1: 1,260; +2: 1,244; +3: 1,248.
  - R-squared: Horizon 0: 0.283; +1: 0.244; +2: 0.266; +3: 0.250.
- Additional seasonal subsample blocks report variant coefficient estimates and show:
  - Temperature coefficients vary in sign and magnitude across seasons and horizons.
  - Several Temperature*extreme heat dummy interaction terms are negative and statistically significant in multiple season–horizon combinations (examples: spring Horizon +2: -0.0442** (0.0190); summer Horizon +2: -0.0496** (0.0212); third-block summer Horizon +3: -0.0246*** (0.00811)).
  - R-squared values across seasonal subsamples often in the 0.5–0.8 range, varying by horizon and season.

*Source: IMF Working Paper — Section 1–4 of "Rising Temperature, Nuanced Effects: Evidence from Seasonal and Sectoral Data" (WP/24/202).*

### Section 1

### Rising Temperature, Nuanced Effects: Evidence from Seasonal and Sectoral Data

### Abstract and central findings
- Uses quarterly temperature and sectoral value-added data for a large sample of advanced economies (AEs) and emerging markets and developing economies (EMDEs).
- For EMDEs:
  - Hotter spring and summer temperatures reduce growth in real value-added of manufacturing, and most significantly, of agriculture.
  - A warmer winter boosts agricultural value-added.
  - A 1-Celsius degree hotter spring reduces year-on-year (yoy) growth in value-added of agriculture by about 0.8 percentage points in the same spring and by more than 1 percentage point in the following summer and fall.
  - A hotter summer also reduces growth in agricultural value-added persistently; a hotter fall does not have a significant effect.
- For AEs:
  - A hotter spring hurts growth in real value-added of all considered sectors: services, manufacturing, and agriculture.
  - A warmer winter boosts agriculture (persistently), but not manufacturing and services.
  - Because agriculture is small in AEs, a warmer winter has a negligible impact on the aggregate economy.
- For both country groups:
  - The negative effect of a hotter spring is larger and more persistent than the positive effect of a warmer winter.
  - At peak, the magnitude of the negative spring aggregate effect is quite similar for EMDEs and AEs.
  - Agriculture is the most affected sector by temperature.
- Time evolution:
  - Adverse impacts of hotter temperatures in advanced economies have accentuated in recent decades; the worsening impacts for AEs are especially significant, large, and consistent across seasons, and were largely driven by services.
  - Worsening impacts for EMDEs are less significant and mostly driven by higher spring temperature.
- Contextual climate facts from the source:
  - The global average temperature is already about 1.1 degree Celsius higher than the pre-industrial level.
  - Example: Highest daily temperature in Washington D.C. in 2021 ranges from 6 degrees Celsius in the winter to the 35 degrees Celsius in the summer; the average annual temperature for Washington D.C. is about 20 degrees Celsius.

### Sectoral and seasonal magnitudes (selected quantitative results and summary statistics)
- Sample and coverage:
  - Real quarterly value-added dataset: 74 countries (30 are AEs and 44 are EMDEs); time coverage spans the 1990-2019 period (pandemic years dropped).
  - Real quarterly GDP: 87 countries (30 are AEs and 57 are EMDEs); time coverage 1990-2019.
  - ERA5 temperature data collected between 1980 and Q2 of 2020 (ERA5 data available to July 9, 2020).
- Table 1: Summary statistics for YoY growth (%) in real quarterly value-added and GDP, 1990-2019 (after winsorization of top and bottom 1%):
  - Agriculture value-added (Number of countries 74): Min -26.418; p1 -19.654; p25 -2.561; Median 2.138; p75 5.722; p99 27.206; Max 36.047.
  - Manufacturing value-added (74): Min -15.397; p1 -12.134; p25 0.105; Median 2.971; p75 6.261; p99 19.132; Max 22.285.
  - Services value-added (74): Min -6.857; p1 -4.573; p25 1.776; Median 3.443; p75 5.551; p99 14.001; Max 17.768.
  - (Agri+Man+Serv) value-added (74): Min -7.659; p1 -5.126; p25 1.692; Median 3.340; p75 5.349; p99 12.285; Max 14.437.
  - GDP (87): Min -9.014; p1 -6.309; p25 1.623; Median 3.528; p75 5.615; p99 12.334; Max 14.100.
- Seasonal definitions used to match temperature with quarterly economic activity:
  - Quarter 1 (January to March) = winter; Quarter 2 (April to June) = spring; Quarter 3 (July to September) = summer; Quarter 4 (October to December) = fall.
  - For Southern hemisphere: winter assigned July to September; spring October to December; summer January to March; fall April to June.
- Temperature construction:
  - Maximum daily temperature from ERA5 averaged across gridcells within a country to construct daily country-level data; daily values averaged across days to generate quarterly seasonal mean of daily maximum temperature; temperature is in Celsius.

### Theoretical motivation and empirical specification
- Two channels of seasonal temperature effects are modeled (based on Dell et al., 2012):
  - Level effect (equation (1)): Y_{s,q} = e^{β_s T_q} A_{s,q}^{α}; β_s captures season-specific and sector-specific level effects of seasonal temperature on sector s value-added.
  - Growth effect (equation (2)): log(A_{s,q}) − log(A_{s,q−4}) = g + λ_s T_q + σ_s T_{q−4}; λ_s and σ_s capture effects of current-season and same-season-last-year temperature on year-on-year seasonal productivity growth.
  - Combined implication for YoY growth (equation (3)): log(Y_{s,q}) − log(Y_{s,q−4}) = αg + (β_s + αλ_s) T_q + (ασ_s − β_s) T_{q−4}; motivates inclusion of current-season temperature T_q and same-season-last-year temperature T_{q−4} in empirical models.
  - Special case with no growth effects (λ_s = σ_s = 0) reduces to YoY value-added growth as a function of year-on-year change in temperature: log(Y_{s,q}) − log(Y_{s,q−4}) = αg + β (T_q − T_{q−4}).

### Data, measurement choices, and limitations highlighted by the authors
- Economic data sources and processing:
  - Real quarterly value-added and GDP from Haver Analytics.
  - Winsorized top and bottom 1% of YoY growth rates to remove large swings.
  - Quarterly value-added data are available for a smaller and, on average, richer set of countries compared to annual datasets; results are expected to represent lower bounds on temperature impacts because poorer and smaller EMDEs (which may be more affected) are underrepresented.
  - No available value-added data for construction in the dataset; construction likely vulnerable to weather shocks but not analyzed here.
- Climate and precipitation data:
  - Temperature and precipitation data from ERA5 (Hersbach et al., 2023) via Google Earth Engine.
  - ERA5 provides hourly data at about 30 km by 30 km grid resolution; analysis uses maximum daily temperature aggregated to quarterly seasonal means.

### Interpretation, policy-relevant implications, and caveats
- Interpretation of short-run vs long-run impacts:
  - The analysis examines short-term effects (up to four quarters) of a temporary shock in temperature.
  - Short-run responses to temperature fluctuations may differ from long-run responses to climate change because of uncertainty in future magnitude of climate change and potential adaptation (e.g., drought-resistant seeds, air-conditioning) that might soften impacts.
  - Conversely, rising temperature could cause irreversible long-run effects (e.g., capital and labor reallocation, emigration) that short-term analysis might understate.
- Policy implications signaled by findings:
  - Season- and sector-specific impacts imply adaptation measures should be tailored to country specificity and sectoral exposure (particularly agriculture and manufacturing in EMDEs; services, manufacturing, and agriculture in AEs for spring effects).
  - The observed worsening of seasonal temperature impacts in recent decades—especially in AEs and driven by services—suggests current adaptation efforts may be insufficient or ineffective to date.

*Prepared by Ha Minh Nguyen and Samuel Pienknagura; IMF Working Paper WP/24/202, September 2024.*

### Section 2

### Section 2

### Data and seasonal temperature patterns
- Daily precipitation data are collected from ERA5, averaged across grids within a country and across the days within a season to generate seasonal average precipitation for a country.
- Summary statistics for temperature, 1980-Q2 2020 (selected entries from Table 2):
  - Number of countries/territories: 219 for all seasons.
  - Temperature Level (Celsius) medians:
    - Winter (Q1): 25.08
    - Spring (Q2): 27.16
    - Summer (Q3): 27.80
    - Fall (Q4): 25.77
  - YoY Change in Temperature (Celsius) medians:
    - Winter (Q1): 0.028
    - Spring (Q2): 0.031
    - Summer (Q3): 0.031
    - Fall (Q4): 0.031
- Average yearly increases in spring temperature (country regressions):
  - Typical median yearly increase: about 0.028 to 0.031 Celsius degrees a year depending on season (median column), translating to about 1.2 Celsius degrees over 40 years.
  - Example country slopes:
    - Vietnam: slope 0.03 (average yearly increase 0.03 Celsius degrees = about 1.2 Celsius degrees in 40 years).
    - Spain: slope 0.072 (average yearly increase 0.072 Celsius degrees = about 2.9 Celsius degrees in 40 years).
- Panel B (Figure 1): Europe shows larger yearly increases in spring temperature relative to many other regions.

### Empirical specification and controls
- Baseline local-projection specification (quarter q, country c, sector s) estimates dynamic impact of quarter q’s temperature on YoY sectoral value-added growth up to h = 1,2,3 quarters ahead:
  - Dependent variable: Δ(VA)_{q+h,c,s} = sector s’ yoy quarterly value-added growth of quarter q+h.
  - Main regressor: Temp_{q,c} = country c’s average temperature in quarter q.
  - Coefficient of interest: β_{s,h} = effect of quarter q’s temperature on YoY growth in value-added of quarter q+h.
- Controls X_{q,c,h} include:
  - Temperature of surrounding quarters: Temp_{q+f,c} for f = −4, −3, −2, −1, 1, ..., h (note: up to four quarters before q and up to h quarters after q).
  - Precipitation: Prec_{q+f,c} for f = −4, −3, −2, −1, 0, 1, .., h (up to four quarters before and up to h quarters after q).
  - Lagged growth in value-added: Δ(VA)_{q+f,c,s} for f = −4, −3, −2, −1.
- Fixed effects:
  - Country fixed effects fe_c to capture country-specific long-run growth and time-invariant aggregation biases.
  - Year-quarter fixed effects fe_q (global shocks for each quarter from 1990 until 2019).
- Regressions run separately for each sector: agriculture, manufacturing and services.
- Note: Short time-series dimension implies focus on impact of changes in seasonal temperature relative to country averages.

### Main findings (seasonal and sectoral effects)
- General:
  - Figure 2 reports dynamic impacts of a 1-Celsius degree higher seasonal temperature on YoY growth of sectoral value-added; left column = AEs, right column = EMDEs; shaded areas = 90% confidence intervals.
  - Distinguishing effects by income group is important to assess differential economic drag and proxy for mediating structural/institutional variables.
- Agriculture:
  - EMDEs:
    - A 1-Celsius degree hotter spring reduces YoY growth in agricultural value-added in the same quarter by about 0.8 percentage points.
    - It also reduces YoY growth in agricultural value-added for the following fall and winter by more than 1 percentage point.
    - A hotter summer has a contemporaneous adverse effect on agricultural growth, with adverse impacts manifesting in the subsequent season (fall) of about 1 percentage points decline in YoY growth.
    - A 1-Celsius degree warmer winter increases growth in agriculture’s value-added, but the impact is less statistically significant.
  - AEs:
    - A 1-Celsius degree hotter spring reduces YoY growth in agricultural value-added in the same season and the following one, but magnitudes are smaller than for EMDEs.
    - A hotter summer reduces agricultural growth in that summer and the following fall.
    - A 1-Celsius degree warmer winter increases growth in agriculture’s value-added quite significantly and the impact is persistent for three quarters (until the following summer).
- Manufacturing:
  - Hotter springs and summers yield lower YoY growth in manufacturing for both AEs and EMDEs, although summer effects are not statistically significant.
  - Effects are substantially smaller than in agriculture and tend to be short-lived.
  - Hotter winters appear to boost manufacturing growth.
  - Larger negative summer effects and larger positive winter effects in EMDEs may reflect worse infrastructure (e.g., less temperature control) and indirect spillovers from agriculture.
- Services:
  - Hotter temperatures have the lowest impact on services.
  - For EMDEs, in most seasons hotter temperatures have a non-significant impact on service YoY growth.
  - For AEs, a hotter spring hurts YoY growth in services in the spring and in subsequent seasons; the impact is small but statistically significant.
  - Possible channels include outdoor-oriented services (gardening, landscaping, outdoor recreation, events), but more granular data are not available for verification.
- Aggregate value-added and GDP:
  - Regression sample for GDP: countries with both value-added and GDP data (27 AEs and 42 EMDEs); sectoral sample is nearly identical (30 AEs and 44 EMDEs).
  - EMDEs:
    - A hotter spring reduces YoY growth in aggregate value-added contemporaneously and in the following season (summer); effect becomes insignificantly different from zero by the following fall.
  - AEs:
    - A hotter spring reduces YoY growth in aggregate value-added in the same spring and persistently across the three subsequent seasons analyzed (very persistent effect).
  - Peak magnitude of the negative spring aggregate effect is quite similar for EMDEs and AEs.
  - Similar results hold for GDP growth: a hotter spring reduces YoY growth in real quarterly GDP in the spring and for subsequent seasons; effects are more persistent for AEs.

### Mechanisms and interpretation
- Agriculture channels:
  - Spring is critical for planting and crop growth (temperature, water, rainfall, pollination). High temperatures cause morpho-anatomical, physiological and biochemical plant changes, heat stress, reduced photosynthesis, increased evaporation and soil moisture loss, increased weeds and pests, and reduced pollination.
  - Hot summers often coincide with harvesting; extreme heat during harvest can disrupt production and plans.
  - These processes explain protracted growth effects, particularly in EMDEs.
- Manufacturing channels:
  - Direct effects via worsened working conditions (reduced labor productivity, higher absenteeism) — especially acute where temperature control infrastructure is limited.
  - Indirect effects via linkages to agriculture (manufacturing inputs tied to agricultural output).
- Services channels:
  - Many service activities occur indoors, reducing sensitivity to outdoor temperature.
  - Outdoor service subsectors in AEs (gardening, landscaping, outdoor recreation, events) may explain the small negative effects of hotter springs in AEs.
- Cross-sectoral seasonal patterns:
  - Hotter winter sometimes boosts agriculture and manufacturing, while hotter spring and summer are typically detrimental to both sectors.

### Impacts by climate zones and income (income × geography interactions)
- Extended specification adds interactions:
  - Temp × Tropics, where tropical countries are defined as countries located south of the Tropic of Cancer and north of the Tropic of Capricorn.
  - Temp × AEs (advanced economies dummy).
- Coefficients of interest: γ_{s,h} (Temp × Tropics) and δ_{s,h} (Temp × AEs).
- Key results:
  - For aggregate value-added growth, AEs do not appear more resilient to higher temperature than EMDEs after controlling for climate zone: AEs*temperature interactions are either not significantly different from zero or in some cases negative.
  - For agriculture specifically, AEs fare better than EMDEs once geographic location (climate zone) is controlled for.

*Source: wpiea2024202-print-pdf - Section 2*

### Section 3

### Section 3

### Interactions by country group and Tropics
- Some interactions AEs*temperature for spring and fall temperature are significant and positive, implying that the same 1-Celsius degree increase in seasonal temperature yields less negative impact on agricultural value-added for AEs than for EMDEs (Panel B of Table 3).
- Due to the small share of agriculture in AE’s economy, the differential impacts for agriculture value-added do not translate to the aggregate value-added.
- The Tropic of Cancer is located at approximately latitude 23°27′ N of the terrestrial equator and the Tropic of Capricorn is located at latitude 23°27′ S.
- Table 3: The dynamic impact of 1-Celsius degree hotter seasonal temperature on aggregate value-added and agricultural value-added (Panels A and B) — econometric results follow equation (7). Robust standard errors in parentheses. * p<0.1, ** p<0.05, *** p<0.01.
  - Table reports seasonal horizons 0, +1, +2, +3 for spring, fall, summer, winter and interactions AEs*temperature and Tropics*temperature, with R-squared and Observations for each specification.

### Changes over time (Decadal analysis)
- Approach:
  - Split sample into three periods: the 1990s (1990-1999), the 2000s (2000-2009) and the 2010s (2010-2019).
  - Interact decadal dummy with temperature (equation (8)).
- Main findings for AEs (Panel A of Table 4; confirmed using real quarterly GDP growth in Table B of Table 4):
  - The adverse impact of a 1-Celsius increase in temperature on aggregate value-added growth is found to be more pronounced in last decade, 2010s, compared to the 1990s.
  - The interaction term 2010s*temperature is largely negative and significant for all seasonal temperature.
  - Results indicate that the increasing impact of temperature on growth is largely driven by services (Panel C of Table 4).
  - Possible interpretations:
    - The effect of a 1-Celsius temperature increase is becoming more severe as the weather gets hotter.
    - The service sector in AEs may be shifting toward more outdoor-oriented industries and hence becoming more subject to temperature (topic for future research).
- Main findings for EMDEs (Table 5):
  - Impacts by decade are more localized to specific seasons.
  - In EMDEs the impact of spring temperature on GDP growth has been more negative since the 2000s.
  - This may be related to EMDEs’ relatively large dependence on agriculture, a sector more susceptible to spring shocks.
  - No statistically significant changes over time are found for other seasons for EMDEs, a pattern that differs relative to AEs.
- Quantitative examples reported in narrative:
  - For EMDEs, a 1-Celsius degree hotter spring reduces yoy growth in agricultural value-added in the same quarter by about 0.8 percentage points and by more than 1 percentage point for the following fall and winter.

### Robustness checks and extensions
- Robustness exercise 1: Include the top and bottom 1% of real value-added growth (no winsorizing).
  - Results largely hold but with larger magnitude and larger standard errors.
  - Hotter spring and summer temperature still hurts agricultural growth in EMDEs (but less so for AEs).
  - For EMDEs, the magnitude of the impact on agriculture is close to 2% in this exercise (larger than baseline), with larger standard errors.
  - A hotter spring temperature hurts growth in services and manufacturing in AEs, consistent with baseline findings.
  - Figure 4: Dynamic impact of a 1-Celsius degree higher seasonal temperature on yoy growth of sectoral value-added (no winsorizing).
    - Panel A: Agriculture; Panel B: Manufacturing; Panel C: Services.
    - Notes: Figure shows estimates 훽_s,h of equation (6). Left column: AEs. Right column: EMDEs. Shaded areas show 90% confidence intervals. All subfigures have the same scale.
- Robustness exercise 2: Control for lags of yoy growth in value-added up to four quarters of all sectors (manufacturing, services, agriculture).
  - Motivation: one sector’s output can be an important input to another (e.g., agri-food industry).
  - After controlling for lags of yoy growth in value-added of all sectors, the quantitative results are unchanged. These additional results are not shown and are available upon request.
- Extreme temperature analysis:
  - Expanded specification (equation (9)) includes 푒푥푡푟푒푚푒 ℎ푒푎푡_q,c, a dummy equal to one if temperature in country c, quarter q is above the 70th percentile of the country-quarter historical distribution.
  - Results (Table 6) show extreme heat can yield larger declines in growth relative to intermediate temperature values in all sectors.
  - Most visible adverse consequences of extreme heat are in agriculture:
    - Extremely hot temperature in the spring, summer and fall leads to a more severe growth slowdown in agriculture.
  - Manufacturing:
    - Extreme heat hurts value-added growth in summers (consistent with literature linking extreme heat to worker absenteeism and lower worker productivity).
  - Services:
    - Extremely hot summer helps value-added growth in services, probably in activities that alleviate heat stress (such as indoor services or beach going).

### Conclusions and policy implications
- Sectoral and seasonal heterogeneity:
  - For EMDEs:
    - Hotter spring and summer temperatures reduce growth in real value-added of manufacturing and, most significantly, of agriculture.
    - A 1-Celsius degree hotter spring reduces yoy growth in agricultural value-added in the same quarter by about 0.8 percentage points and by more than 1 percentage point for the following fall and winter.
    - A warmer winter boosts agricultural activity.
  - For AEs:
    - A hotter spring hurts growth in real value-added of services, manufacturing and agriculture.
  - For both country groups:
    - The negative effect of a hotter spring is larger and more persistent than the positive effect of a warmer winter.
    - For EMDEs, the more negative spring effect is driven more by agriculture and manufacturing; for AEs, it is driven by agriculture, manufacturing and services.
- Broader implications:
  - Heterogeneous impacts across sectors imply timing of climate shocks and sectoral composition of the economy matter when gauging expected impacts of climate change.
  - Larger impacts on agriculture—where a large share of low-income workers are employed—point to potentially adverse consequences of climate change on poverty and inequality.
  - The impact of temperature on economic activity is mainly driven by extreme heat; hotter seasons, especially above-mean temperatures, are likely to have particularly negative impacts on economic activity.
  - The adverse impacts of hotter temperatures in advanced economies and, to a lesser extent, EMDEs have accentuated in recent decades; in AEs this trend is largely driven by services.
- Policy recommendations and emphasis:
  - Importance of investments in adaptation that help improve resilience to changing climate conditions.
  - Importance of adjusting economic structure to reduce susceptibility to temperature over time, especially in service-intensive advanced economies.

*Italic line: Source: IMF Working Paper — Section 3 of "Rising Temperature, Nuanced Effects: Evidence from Seasonal and Sectoral Data" (figures, tables and equations as presented).*

### Section 4

### Section 4

### Estimation details and significance notation
- Robust standard errors in parentheses.
- Significance: * p<0.1, ** p<0.05, *** p<0.01.

### Spring — key coefficients, horizons 0 to +3
- Temperature:
  - Horizon 0: -0.333 (0.277)
  - Horizon +1: -0.561* (0.308)
  - Horizon +2: 0.0625 (0.300)
  - Horizon +3: 0.880*** (0.308)
- Temperature*extreme heat dummy:
  - Horizon 0: -0.00533 (0.0194)
  - Horizon +1: -0.0215 (0.0176)
  - Horizon +2: -0.0442** (0.0190)
  - Horizon +3: -0.0520** (0.0242)
- Constant:
  - Horizon 0: 5.026 (9.703)
  - Horizon +1: 0.551 (11.95)
  - Horizon +2: 11.28 (11.52)
  - Horizon +3: -37.83** (15.92)
- Observations:
  - Horizon 0: 1,256
  - Horizon +1: 1,243
  - Horizon +2: 1,248
  - Horizon +3: 1,192
- R-squared:
  - Horizon 0: 0.627
  - Horizon +1: 0.504
  - Horizon +2: 0.462
  - Horizon +3: 0.227

### Fall — key coefficients, horizons 0 to +3
- Temperature:
  - Horizon 0: 0.112 (0.153)
  - Horizon +1: 0.550** (0.229)
  - Horizon +2: 0.451 (0.282)
  - Horizon +3: 0.782** (0.331)
- Temperature*extreme heat dummy:
  - Horizon 0: -0.00681 (0.0192)
  - Horizon +1: -0.0286 (0.0244)
  - Horizon +2: -0.0571* (0.0289)
  - Horizon +3: -0.0401 (0.0281)
- Constant:
  - Horizon 0: 15.47* (8.339)
  - Horizon +1: -38.27*** (10.06)
  - Horizon +2: -23.62 (16.45)
  - Horizon +3: -19.12 (15.15)
- Observations:
  - Horizon 0: 1,197
  - Horizon +1: 1,196
  - Horizon +2: 1,195
  - Horizon +3: 1,178
- R-squared:
  - Horizon 0: 0.599
  - Horizon +1: 0.250
  - Horizon +2: 0.243
  - Horizon +3: 0.273

### Summer — key coefficients, horizons 0 to +3
- Temperature:
  - Horizon 0: -0.375 (0.282)
  - Horizon +1: -0.818** (0.312)
  - Horizon +2: 0.735 (0.471)
  - Horizon +3: 1.241*** (0.467)
- Temperature*extreme heat dummy:
  - Horizon 0: -0.00816 (0.0172)
  - Horizon +1: -0.0119 (0.0243)
  - Horizon +2: -0.0496** (0.0212)
  - Horizon +3: -0.0484** (0.0220)
- Constant:
  - Horizon 0: 7.321 (11.30)
  - Horizon +1: 19.98* (11.30)
  - Horizon +2: -42.76** (16.90)
  - Horizon +3: -33.31 (22.97)
- Observations:
  - Horizon 0: 1,254
  - Horizon +1: 1,257
  - Horizon +2: 1,199
  - Horizon +3: 1,197
- R-squared:
  - Horizon 0: 0.632
  - Horizon +1: 0.522
  - Horizon +2: 0.239
  - Horizon +3: 0.245

### Winter — key coefficients, horizons 0 to +3
- Temperature:
  - Horizon 0: 0.611*** (0.166)
  - Horizon +1: 0.364* (0.193)
  - Horizon +2: 0.715*** (0.207)
  - Horizon +3: 0.0360 (0.193)
- Temperature*extreme heat dummy:
  - Horizon 0: -0.0159 (0.0226)
  - Horizon +1: -0.0154 (0.0247)
  - Horizon +2: -0.00444 (0.0251)
  - Horizon +3: 0.0212 (0.0264)
- Constant:
  - Horizon 0: -21.68* (11.11)
  - Horizon +1: -18.66 (14.28)
  - Horizon +2: -5.287 (12.55)
  - Horizon +3: 13.44 (10.88)
- Observations:
  - Horizon 0: 1,263
  - Horizon +1: 1,260
  - Horizon +2: 1,244
  - Horizon +3: 1,248
- R-squared:
  - Horizon 0: 0.283
  - Horizon +1: 0.244
  - Horizon +2: 0.266
  - Horizon +3: 0.250

### Seasonal subsamples — additional horizon-specific estimates (spring, fall, summer, winter)
- Spring (second block) — Temperature:
  - Horizon 0: -0.239 (0.156)
  - Horizon +1: -0.240 (0.176)
  - Horizon +2: -0.123 (0.178)
  - Horizon +3: -0.221 (0.206)
- Spring — Temperature*extreme heat dummy:
  - Horizon 0: -0.000335 (0.0121)
  - Horizon +1: 0.00681 (0.0144)
  - Horizon +2: 0.000259 (0.0162)
  - Horizon +3: 0.00246 (0.0182)
- Spring — Constant:
  - Horizon 0: 6.892 (6.126)
  - Horizon +1: 13.86 (9.167)
  - Horizon +2: 18.55* (10.02)
  - Horizon +3: 7.554 (10.85)
- Spring — Observations:
  - Horizon 0: 1,256; +1: 1,254; +2: 1,249; +3: 1,169
- Spring — R-squared:
  - Horizon 0: 0.716; +1: 0.581; +2: 0.490; +3: 0.419

- Fall (second block) — Temperature:
  - Horizon 0: 0.0713 (0.0929)
  - Horizon +1: -0.0400 (0.127)
  - Horizon +2: -0.0807 (0.138)
  - Horizon +3: 0.0726 (0.161)
- Fall — Temperature*extreme heat dummy:
  - Horizon 0: -0.00793 (0.0136)
  - Horizon +1: -0.00991 (0.0164)
  - Horizon +2: -0.00161 (0.0188)
  - Horizon +3: -0.0210 (0.0188)
- Fall — Constant:
  - Horizon 0: -8.805** (3.680)
  - Horizon +1: -5.626 (5.942)
  - Horizon +2: 9.015 (8.156)
  - Horizon +3: 13.70 (9.680)
- Fall — Observations:
  - Horizon 0: 1,205; +1: 1,189; +2: 1,197; +3: 1,192
- Fall — R-squared:
  - Horizon 0: 0.724; +1: 0.554; +2: 0.501; +3: 0.409

- Summer (second block) — Temperature:
  - Horizon 0: -0.0350 (0.180)
  - Horizon +1: 0.257 (0.238)
  - Horizon +2: 0.347 (0.302)
  - Horizon +3: 0.469 (0.306)
- Summer — Temperature*extreme heat dummy:
  - Horizon 0: -0.0176* (0.00999)
  - Horizon +1: -0.0182 (0.0139)
  - Horizon +2: -0.0237 (0.0170)
  - Horizon +3: -0.0321* (0.0186)
- Summer — Constant:
  - Horizon 0: 7.027 (6.420)
  - Horizon +1: 6.680 (8.407)
  - Horizon +2: -4.126 (9.459)
  - Horizon +3: 8.679 (11.94)
- Summer — Observations:
  - Horizon 0: 1,266; +1: 1,261; +2: 1,181; +3: 1,189
- Summer — R-squared:
  - Horizon 0: 0.744; +1: 0.588; +2: 0.486; +3: 0.447

- Winter (second block) — Temperature:
  - Horizon 0: 0.154 (0.133)
  - Horizon +1: 0.0471 (0.124)
  - Horizon +2: 0.00818 (0.135)
  - Horizon +3: -0.0723 (0.145)
- Winter — Temperature*extreme heat dummy:
  - Horizon 0: 0.0171 (0.0155)
  - Horizon +1: 0.00925 (0.0150)
  - Horizon +2: -0.000223 (0.0193)
  - Horizon +3: 0.00496 (0.0207)
- Winter — Constant:
  - Horizon 0: 7.675 (5.980)
  - Horizon +1: 11.91* (6.694)
  - Horizon +2: 21.08** (10.19)
  - Horizon +3: 22.70** (9.739)
- Winter — Observations:
  - Horizon 0: 1,253; +1: 1,261; +2: 1,254; +3: 1,256
- Winter — R-squared:
  - Horizon 0: 0.577; +1: 0.531; +2: 0.442; +3: 0.401

### Additional seasonal blocks — third set of estimates (spring, fall, summer, winter)
- Spring (third block) — Temperature:
  - Horizon 0: -0.0680 (0.0913)
  - Horizon +1: -0.128 (0.0805)
  - Horizon +2: -0.241** (0.0958)
  - Horizon +3: -0.244** (0.112)
- Spring — Temperature*extreme heat dummy:
  - Horizon 0: 7.94e-05 (0.00658)
  - Horizon +1: 0.00158 (0.00696)
  - Horizon +2: 0.00412 (0.00691)
  - Horizon +3: 0.0122 (0.00835)
- Spring — Constant:
  - Horizon 0: 0.847 (2.924)
  - Horizon +1: 0.477 (3.872)
  - Horizon +2: 1.928 (4.681)
  - Horizon +3: 4.528 (5.966)
- Spring — Observations:
  - Horizon 0: 1,282; +1: 1,278; +2: 1,273; +3: 1,209
- Spring — R-squared:
  - Horizon 0: 0.740; +1: 0.686; +2: 0.616; +3: 0.531

- Fall (third block) — Temperature:
  - Horizon 0: 0.0859 (0.0774)
  - Horizon +1: 0.00183 (0.0907)
  - Horizon +2: 0.0393 (0.0822)
  - Horizon +3: 0.0530 (0.0932)
- Fall — Temperature*extreme heat dummy:
  - Horizon 0: -0.00362 (0.00797)
  - Horizon +1: -0.0101 (0.00751)
  - Horizon +2: -0.0126 (0.00934)
  - Horizon +3: -0.00899 (0.0102)
- Fall — Constant:
  - Horizon 0: 3.870* (2.260)
  - Horizon +1: 0.376 (4.118)
  - Horizon +2: 5.071 (4.151)
  - Horizon +3: 1.902 (4.954)
- Fall — Observations:
  - Horizon 0: 1,221; +1: 1,220; +2: 1,221; +3: 1,209
- Fall — R-squared:
  - Horizon 0: 0.740; +1: 0.639; +2: 0.595; +3: 0.522

- Summer (third block) — Temperature:
  - Horizon 0: -0.0175 (0.103)
  - Horizon +1: 0.130 (0.133)
  - Horizon +2: 0.236 (0.144)
  - Horizon +3: 0.368*** (0.128)
- Summer — Temperature*extreme heat dummy:
  - Horizon 0: 0.00517 (0.00654)
  - Horizon +1: -0.0141* (0.00713)
  - Horizon +2: -0.0158* (0.00911)
  - Horizon +3: -0.0246*** (0.00811)
- Summer — Constant:
  - Horizon 0: -1.265 (2.933)
  - Horizon +1: -3.771 (3.291)
  - Horizon +2: -4.741 (5.265)
  - Horizon +3: 2.643 (4.692)
- Summer — Observations:
  - Horizon 0: 1,287; +1: 1,281; +2: 1,219; +3: 1,220
- Summer — R-squared:
  - Horizon 0: 0.786; +1: 0.670; +2: 0.571; +3: 0.560

- Winter (third block) — Temperature:
  - Horizon 0: 0.00377 (0.0806)
  - Horizon +1: -0.0798 (0.0714)
  - Horizon +2: 0.0242 (0.0637)
  - Horizon +3: 0.0173 (0.0674)
- Winter — Temperature*extreme heat dummy:
  - Horizon 0: 0.00501 (0.00746)
  - Horizon +1: 0.000279 (0.00792)
  - Horizon +2: 0.00679 (0.00900)
  - Horizon +3: 0.0121 (0.0107)
- Winter — Constant:
  - Horizon 0: 3.495 (4.005)
  - Horizon +1: 5.934 (3.594)
  - Horizon +2: 2.354 (4.137)
  - Horizon +3: 3.688 (4.448)
- Winter — Observations:
  - Horizon 0: 1,283; +1: 1,283; +2: 1,271; +3: 1,265
- Winter — R-squared:
  - Horizon 0: 0.652; +1: 0.625; +2: 0.568; +3: 0.531

### Empirical patterns and interpretation (as indicated by estimates)
- Temperature coefficients vary in sign and magnitude across seasons and forecast horizons; several estimates are statistically significant at conventional levels (examples: spring Horizon +3: 0.880*** (0.308); summer Horizon +3: 1.241*** (0.467); winter Horizon 0: 0.611*** (0.166)).
- Interaction terms Temperature*extreme heat dummy are often negative and reach significance in multiple season–horizon combinations (examples: spring Horizon +2: -0.0442** (0.0190); summer Horizon +2: -0.0496** (0.0212); summer Horizon +3: -0.0484** (0.0220); third-block summer Horizon +3: -0.0246*** (0.00811)).
- R-squared values indicate higher model fit in several seasonal subsamples (many values in the 0.5–0.8 range), but fit varies by horizon and season.

*Working Paper No. WP/2024/202*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024202-print-pdf.pdf_