## wpiea2025220-source-pdf

## Source details

**Canonical URL:** [wpiea2025220-source-pdf](https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025220-source-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2025/english/wpiea2025220-source-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2025/english/wpiea2025220-source-pdf.pdf.json)

---

### Overview
- Trade policy activity appears to have intensified in recent years, with governments deploying multiple instruments—tariffs, non-tariff measures, subsidies—at increasing frequency, often simultaneously restricting and facilitating trade.
- Existing indicators capture important but partial aspects of trade policy, leaving the dynamics of global policy activity inadequately captured.
- This paper constructs the Trade Policy Activity (TPA) Index, a monthly indicator of global trade policy dynamics covering 197 countries and territories since the Global Financial Crisis.

### TPA Index construction and purpose
- Data sources:
  - WTO Trade Monitoring Database (TMDB).
  - Global Trade Alert (GTA).
- Method:
  - Apply a Dynamic Factor Model (DFM) to extract the common component of policy variation across measure categories and data sources.
  - Block structure separates the common global signal from category-specific dynamics (facilitating vs other measures).
  - Endogenous weighting: the model determines the weight of each input series from the covariance structure rather than imposing weights ex ante.
- Key features:
  - Monthly frequency.
  - Global scope (197 countries and territories).
  - Tracks policy changes (activity) rather than the level of restrictiveness.
  - Distinguishes long-term pronounced shifts from cyclical fluctuations.
  - Permits disaggregation by country groups.

### Data sources and data construction details
- Databases and creation dates:
  - TMDB created in October 2008.
  - GTA created in June 2009.
- Coverage differences:
  - GTA tracks a broader set (including subsidies and measures applied to specific firms or subnational bodies) than TMDB.
  - TMDB relies mainly on official government sources and records implemented measures only; GTA includes policy announcements and unofficial sources.
  - TMDB data is released twice a year; GTA includes ongoing updates.
- Input variables for the DFM:
  - Monthly counts of new trade-policy measures introduced globally (extensive margin).
  - Average number of affected products (HS 6-digit) by measure category and data source (extent of application).
- GTA discovery-bias correction:
  - Implement a consistent 12-month discovery window following Evenett and Fritz (2018): a measure is included only if it was discovered within 12 months of its announcement date.
  - On average, 78% of measures are discovered within this window, with no clear change over time.
  - Empirical statistic: Figure A1.2 reports Mean = 77.9%; G20 mean = 77.3%; non-G20 mean = 80.0%.
- Sample period for baseline index:
  - January 2010 to October 2025 when both databases have full coverage.
  - Note: TMDB provides month of implementation; GTA provides month of announcement.

### Model specification and identification
- Core model (Equation 1): y*t = φ(t) (Nonlinear Trend) + Λg fg,t (Global Factor) + Λl fl,t (Local Factors) + εt (Idiosyncratic Noise).
- Trend and factors:
  - φ(t) = [φ1(t),...,φn(t)]′ deterministic nonlinear time trends.
  - fg,t ∈ R is a global factor capturing common co-movements across all trade variables.
  - fl,t ∈ R^2 are two local factors capturing idiosyncratic co-movements between facilitating and other measures.
  - Factors are mean-independent of t and stationary.
- Block structure on loadings Λ (Equation 4) separates facilitating and other measures.
- Normalization:
  - Global factor adjusted so average value = 0 over January 2010–December 2011; divided by absolute value of the average global factor in the same period to set baseline = 0.

### Estimation procedure and diagnostics
- Two-step estimation:
  1. Remove deterministic nonlinear trend using local quadratic regression weighted by inverse distance between points.
  2. Construct joint likelihood of (εt, ut) following Bańbura and Modugno (2014) and maximize via expectation-maximization (EM) algorithm combining Kalman smoother (E-step) with multivariate regressions (M-step).
- Assumptions:
  - εt independent of ut.
  - εt ∼ N(0,R) with R diagonal.
- Practical diagnostics and properties:
  - Seasonality removed using X-13ARIMA-SEATS.
  - Missing values account for about 0.2% of total sample size.
  - Stationarity checks: KPSS does not reject stationarity and ADF rejects the unit root in all cases.
  - Estimated autoregressive coefficient on the global factor = 0.6 (approximately 60% persistence month-to-month).
  - Rolling estimation with 72-observation windows produces an index within the 90% confidence intervals of the baseline.
  - Factor trend recovery: Φf = (Λ′Λ)−1(Λ′Φ) and f*t = ft + Φf,t (Equation 6).
  - Estimated factor trends revert after about 6 months in simulation of trend reversion.

### Baseline TPA Index — main findings and temporal features
- Definition: TPA Index is the normalized global factor (baseline value 0 for January 2010–December 2011); positive values indicate heightened activity relative to the baseline.
- Distinct peaks coincide with:
  - the rise in trade tensions in 2025;
  - the war in Ukraine (2022);
  - the COVID-19 pandemic (2020);
  - the U.S.–China tariff increases (2018–2019).
- Additional earlier peaks:
  - 2013 (WTO Trade Facilitation Agreement conclusion).
  - 2016 (possible responses to China’s 2015 currency devaluation, surge in trade remedy investigations, Brexit uncertainty).
- Trend:
  - Rising trend that accelerates from 2020 onward with the 90% confidence interval remaining above zero from 2022 onward.
- Relationship to other indices (cointegration and fit):
  - Engle-Granger two-step procedure rejects null of no cointegration at the 10% level between TPA and each compared index.
  - Cointegrating regression R2 with Fernández-Villaverde et al. (2024) = 0.51.
  - R2 with Caldara et al. (2020) = 0.14.
  - R2 with Caldara and Iacoviello (2022) = 0.18.
  - R2 with Ahir et al. (2022) = 0.03.
- Interpretation: index captures both short-term spikes linked to events and a secular rise tied to expanded use of trade policy for trade and non-trade objectives and geoeconomic fragmentation.

### Role of measure categories (facilitating, restrictive, other)
- Classification:
  - “Facilitating”: TMDB ‘facilitating’ and GTA ‘green’ measures; actions that reduce restrictiveness or ease trade, including rollbacks.
  - “Other”: trade-restrictive measures and other potentially trade-distortive measures (TMDB ‘remedial’ and ‘other’, GTA ‘red’ and ‘amber’).
  - Within “Other”, distinction made between “Restrictive Measures” (tariffs and non-tariff measures under MAST chapters E, C, I and selected subcomponents of P) and “Remaining Other Measures” (largely subsidies and other distortions).
- Counterfactual approach:
  - Re-estimate global factor using baseline loadings for one category at a time with normalization steps: average value = zero over January 2010–December 2011; cyclical component rescaled to match standard deviation of baseline global factor over January 2010–December 2019; divide by absolute average TPA over January 2010–December 2011 for comparability.
- Findings:
  - Both facilitating and other measures show broadly rising trajectories, but “other measures” increasingly shape the global factor, with divergence widening over 2022–2025.
  - Crisis episodes (COVID-19, war in Ukraine): facilitating and other measures respond in tandem (local factors near zero), with global factor capturing common response.
  - 2018–2019 and 2025 trade tensions: local factors diverge—other measures accelerate while facilitating measures decelerate relative to the common signal.
  - Remaining other measures show a broad-based increase over the sample (consistent with rising industrial policy).
  - The 2025 spike is explained predominantly by restrictive measures (tariffs and export controls).
  - Facilitating measures show a distinct rise following WTO TFA entry into force in 2017.

### Dynamics across country groups and weighting
- Group-specific estimation:
  - Global factor estimated separately for G20 and non-G20 country groups.
- Findings by group:
  - Both G20 and non-G20 record positive trends; increase begins earlier for G20.
  - Non-G20 recorded larger spikes during the COVID-19 pandemic.
  - 2017–2018 and 2025 peaks are more pronounced for G20 economies.
  - Within-group decomposition:
    - Remaining other measures (largely subsidies) increase markedly in both G20 and non-G20.
    - Among G20, restrictive measures diverge sharply in 2018–2019 and especially 2025.
    - Among non-G20, remaining other measures spike during COVID-19; 2025 increase in restrictive measures is less pronounced; facilitating measures subdued from 2022 onward.
  - Caveat: patterns could be partially affected by differences in coverage of underlying data across country groups.
- GDP-weighted index:
  - Construction uses countries’ pre-sample average share in global GDP (2005–2009) as time-invariant weights.
  - When weighted by economic size, rise in activity is less pronounced around COVID-19; peaks in 2018–2019 and early 2025 are more prominent.
  - Excluding U.S. and China confirms their important contribution, especially in 2025, but also shows broad-based acceleration in restrictive measures since 2022.

### Robustness checks and sensitivity
- Model specification:
  - Alternative specification with a single global factor and no block restriction yields similar dynamics though higher values post-2019.
- Trend extraction:
  - Tested different weight functions and trend-filtering method of Tibshirani (2014); global factor shape robust to method.
  - Baseline trend extraction uses a weighted quadratic regression (local quadratic all); alternatives (local quadratic with 33% weights, quadratic least-squares, Tibshirani trend filter) produce qualitatively similar trends.
- Rolling-window estimation and extensions:
  - Rolling-window estimation with 72 observations produces index within the 90% confidence intervals of the baseline.
  - Baseline update covers through October 2025; extensions through January 2026 use rolling-window approaches with bandwidth 0.75 (~144 observations) or alternative 190-observation windows when adding months one at a time while dropping the oldest.
  - For months with incomplete discovery (November 2025–January 2026), month-specific adjustment factors can be applied assuming stable discovery patterns across time and country groups.
- Input and sample variations:
  - Tests adding number of implementing countries; counts-only vs counts+product coverage; inclusion of broader GTA measures (capital controls, FDI restrictions) — overall dynamics consistent though peak magnitudes vary with input dimension.
  - Re-estimation with alternative end dates (October 2020 and October 2022) yields consistent loadings and dynamics.
- Missingness:
  - Estimation approach accommodates missingness (~0.2%).
- Additional note:
  - A version holding trends and loadings constant at baseline produces nearly identical historical estimates; divergence at sample end reflects informational gains from re-estimation.

### Dynamic factor model inputs and normalization details
- Time-series variables used (by source and category):
  - GTA:
    - Total number of measures — Facilitating
    - Total number of measures — Other
    - Average number of products — Facilitating
    - Average number of products — Other
  - TMDB:
    - Total number of measures — Facilitating
    - Total number of measures — Other
    - Average number of products — Facilitating
    - Average number of products — Other
- Block structure and baseline model:
  - Baseline model uses one global factor and two local factors (Facilitating and Other).
  - Factors normalized relative to January 2010–December 2011 reference period (baseline = 0).
  - Final reported factor is smoothed using a three-month moving average for clearer presentation.
- Loadings and contributions:
  - GTA series exhibit larger loadings on the global factor relative to TMDB series, reflecting greater variation in GTA inputs.
  - Decomposition of baseline trend into GTA-only and TMDB-only contributions is exactly additive; GTA-only and TMDB-only trends sum to the baseline trend at every point while sharing the same cyclical component.

### Figures A2.12–A2.14 — alternative weighting, exclusion of U.S. and China, and rolling-window
- Alternative approach accounting for economic size by policy category (Figure A2.12):
  - Countries’ shares in global GDP (average during 2005-2009) used as weights.
  - Normalization and scaling:
    - Average equals zero over January 2010–December 2011.
    - Cyclical component rescaled to match standard deviation of baseline global factor over January 2010–December 2019.
    - Divide by absolute value of baseline global factor’s average over January 2010–December 2011.
  - Average used for GDP weights: 2005-2009.
  - Normalization window: January 2010–December 2011.
  - Rescaling window for standard deviation: January 2010–December 2019.
- TPA index outside of U.S. and China (Figure A2.13):
  - Dashed lines represent estimates excluding the United States and China.
  - Same normalization and rescaling procedure as above.
- Rolling-window estimation of the global factor (Figure A2.14):
  - Rolling samples of 72 observations.
  - Rolling estimate lies within the 90% confidence intervals of the baseline (confidence interval reported: 90%).
  - Rolling-window sample size: 72 observations.

### Contributions, applications, and conclusion
- Three main contributions:
  1. Identify distinct surges in global trade policy activity amid a rising trend accelerating from 2020 onward, providing monthly evidence across a wide set of instruments and countries.
  2. Advance trade policy measurement by tracking dynamics of policy activity (monthly, global scope), covering both facilitating and restrictive measures, complementing welfare-based indices and news-based uncertainty indices.
  3. Extend Dynamic Factor Model methods to trade policy analysis, exploiting covariation across policy categories and data sources to recover a latent global factor representing common policy dynamics.
- Potential empirical uses and operational applications:
  - Serve as a control variable to address omitted variable bias in studies of specific policy changes.
  - Use monthly frequency for event studies and local projections to study dynamic effects of trade policy shocks.
  - Examine how aggregate policy environment moderates international transmission of shocks (trade, investment reallocation).
  - Revisit determinants of trade policy and its cyclicality.
  - Operational use: robust to rolling-window estimation and suitable for periodic updating and nowcasting; Appendix demonstrates extension through January 2026 under alternative assumptions.

*Source: Introduction and related sections from wpiea2025220-source-pdf*

### Introduction

### Introduction

### Overview
- Trade policy activity appears to have intensified in recent years, with governments deploying multiple instruments—tariffs, non-tariff measures, subsidies—at increasing frequency, often simultaneously restricting and facilitating trade.
- Existing indicators capture important but partial aspects of trade policy, leaving the dynamics of global policy activity inadequately captured.
- This paper constructs the Trade Policy Activity (TPA) Index, a monthly indicator of global trade policy dynamics covering 197 countries and territories since the Global Financial Crisis.

### TPA Index construction and purpose
- Data sources: WTO Trade Monitoring Database (TMDB) and Global Trade Alert (GTA).
- Method: apply a Dynamic Factor Model (DFM) to extract the common component of policy variation across measure categories and data sources.
- Key features of the index:
  - Monthly frequency.
  - Global scope (197 countries and territories).
  - Tracks policy changes (activity) rather than the level of restrictiveness.
  - Distinguishes long-term pronounced shifts from cyclical fluctuations.
  - Block structure separates the common global signal from category-specific dynamics (facilitating vs other measures).
  - Permits disaggregation by country groups.

### Key findings and dynamics identified
- The TPA index identifies distinct peaks coinciding with:
  - the escalation of US-China tariffs (2018-2019),
  - pandemic-related trade measures (2020),
  - the war in Ukraine (2022),
  - increased trade tensions in 2025.
- The index documents a rising trend that accelerates from 2020 onward.
- Co-movement:
  - Close co-movement between facilitating and restrictive measures at the global level, consistent with crisis-driven policy bundling and strategic interdependence.
  - Certain episodes show category-specific dynamics beyond the common signal, notably the asymmetric response of restrictive and facilitating measures during the 2025 trade tensions.
  - Facilitating measures display a distinct rise following the WTO Trade Facilitation Agreement that is largely absent in other categories.
- Country-group heterogeneity:
  - Larger economies (G20) exhibit a more persistent upward trend in policy activity.
  - This trend is not solely explained by countries dominating the headlines.
- Robustness: findings hold under alternative specifications and samples, various approaches to trend extraction, and exclusion of dominant economies.

### Contributions to literature and measurement
- Three main contributions:
  1. Identify distinct surges in global trade policy activity amid a rising trend accelerating from 2020 onward, providing monthly evidence across a wide set of instruments and countries.
  2. Advance trade policy measurement by tracking dynamics of policy activity (monthly, global scope), covering both facilitating and restrictive measures, complementing welfare-based indices and news-based uncertainty indices.
  3. Extend Dynamic Factor Model methods to trade policy analysis, exploiting covariation across policy categories and data sources to recover a latent global factor representing common policy dynamics.
- Comparisons with existing measures:
  - Complements welfare-based indices (Anderson and Neary (2005); Kee et al. (2009)) by tracking activity rather than level.
  - Complements uncertainty and risk indices constructed from news coverage (Caldara et al. (2020); Caldara and Iacoviello (2022); Ahir et al. (2022)) by capturing underlying policy changes, including those not attracting public attention.
  - Complements geopolitical fragmentation indices (Fernández-Villaverde et al. (2024)) by focusing specifically on trade policy changes.
  - Differs from other annual-frequency approaches (Cerdeiro and Nam (2018); Furceri et al. (2022)) by providing monthly frequency and extracting common dynamics across instruments.

### Data sources
- Two complementary databases used:
  - WTO Trade Monitoring Database (TMDB): records trade policy measures implemented by WTO Members and Observers identified through formal WTO channels and official sources (WTO, 2024).
  - Global Trade Alert (GTA): compiles announced and implemented measures from a wide set of publicly available sources, including press articles (Global Trade Alert, 2022).
- TMDB and GTA were created after the Global Financial Crisis:
  - TMDB was created in October 2008 and GTA in June 2009.
- Differences in coverage:
  - Scope: GTA tracks a broader set (including subsidies and measures applied to specific firms or subnational bodies) than TMDB.
  - Sources: TMDB relies mainly on official government sources and records implemented measures only; GTA additionally scouts unofficial sources and includes policy announcements.
  - Release frequency: TMDB data is released twice a year, adding new measures for the period; GTA includes ongoing updates.
- Coverage of measures:
  - TMDB includes import and export restrictions (tariffs, quantitative restrictions and other taxes), trade remedies (anti-dumping, countervailing and safeguard measures), customs-related procedures, trade-related investment measures (local content requirements) and other trade measures.
  - GTA covers a wider range that can affect trade, including subsidies (financial and in-kind grants, state loans or state aid), capital controls and exchange rate policy, foreign direct investment (FDI) restrictions, intellectual property, and other measures.
- Scope for index construction:
  - Baseline Index considers only trade-policy related measures (see Appendix Table A1.1 in source).
  - Sample period: January 2010 to October 2025 when both databases have full coverage.
  - Note: TMDB provides month of implementation; GTA provides month of announcement.

### Data construction details
- Input variables:
  - Monthly counts of new trade-policy measures introduced globally (captures extensive margin).
  - Average number of affected products (HS 6-digit) by measure category and data source (gauges extent of application).
- GTA discovery bias correction:
  - Implement a consistent 12-month discovery window following Evenett and Fritz (2018): a measure is included only if it was discovered within 12 months of its announcement date.
  - On average, 78% of measures are discovered within this window, with no clear change over time.
- Classification into categories:
  - Policies are divided into “Facilitating” and “Other”.
    - “Facilitating”: actions that clearly reduce restrictiveness or ease trade, including rollbacks of prior restrictions (TMDB ‘facilitating’ and GTA ‘green’ measures).
    - “Other”: trade-restrictive measures and other potentially trade-distortive measures, such as tariffs, bans, quotas, subsidies (TMDB ‘remedial’ and ‘other’, GTA ‘red’ and ‘amber’). MAST classification is used to further differentiate within the other category.
- Robustness checks include alternative sets of measures and additional variable specifications (e.g., including implementing countries).

### Methodology
- Dynamic Factor Model (DFM) approach:
  - Extract common movements from a high-dimensional set of deseasonalized time series.
  - Seasonality removed using X-13ARIMA-SEATS (U.S. Census Bureau, 2019).
  - Work with an n×1 vector of deseasonalized time series y* t = [y* 1t , ..., y* nt ]′ , where i = 1,...,n indicates individual time series (see Appendix Table A1.2 in source for list).
  - Model explicitly accounts for trending behavior and structural relationships between different categories of measures.
  - Block structure: distinguishes facilitating from other measures, allowing the model to capture both the shared global signal and category-specific variation.
  - Endogenous weighting: the model determines the weight of each input series from the covariance structure rather than imposing weights ex ante.

### Paper structure
- Section I describes the data sources and construction.
- Section II presents the dynamic factor model.
- Section III reports the baseline index, examines dynamics by measure category and country group, and presents robustness checks.
- Section IV concludes.
- A Supplemental Appendix (Appendix) provides further data description and additional results.

*Source: Introduction from wpiea2025220-source-pdf - Introduction*

### 2025. For a version through January 2026, based on additional assumptions, see Section A1.6 in Appendix A1.

### wpiea2025220-source-pdf - 2025. For a version through January 2026, based on additional assumptions, see Section A1.6 in Appendix A1.

### Model specification and identification
- Core model (Equation 1): y*t = φ(t) (Nonlinear Trend) + Λg fg,t (Global Factor) + Λl fl,t (Local Factors) + εt (Idiosyncratic Noise).
- Trend: φ(t) = [φ1(t),...,φn(t)]′ is an n×1 vector of deterministic nonlinear time trends.
- Factors:
  - fg,t ∈ R is a global factor capturing common co-movements across all trade variables.
  - fl,t ∈ R^2 are two local factors capturing idiosyncratic co-movements between facilitating and other measures.
  - Factors are mean-independent of t and stationary (no deterministic or stochastic trend).
- Factor dynamics (Equation 2): ft = A ft−1 + Q ut, where ut ∼ N(0,I3) and I3 is a 3×3 identity matrix.
- Trend-cycle decomposition: f*t = ft + Φf(t) and y*t = Λ f*t + εt = Λ Φf(t) + Λ ft + εt (Equation 3), implying φ(t) = Λ Φf(t).
- Block structure (Equation 4) on loadings Λ to separate cross-category commonality and within-category idiosyncrasy:
  - Λ = [Λg Λl] = [ [Λg,f Λf,f 0]; [Λg,o 0 Λo,o] ] (arranged by facilitating and other measures).
- Normalization: global factor adjusted so average value = 0 over January 2010–December 2011; divided by absolute value of the average global factor in the same period to set baseline = 0.

### Estimation procedure and diagnostics
- Two-step estimation:
  1. Remove deterministic nonlinear trend using local quadratic regression weighted by inverse distance between points (local regression).
  2. Construct joint likelihood of (εt, ut) following Bańbura and Modugno (2014) and maximize via the expectation-maximization (EM) algorithm combining Kalman smoother (E-step) with multivariate regressions (M-step).
- Assumptions for likelihood:
  - εt independent of ut.
  - εt ∼ N(0,R) with R diagonal.
- Practical details and properties:
  - Detrended series yt = [y1t,...,ynt]′ satisfy yt = Λ ft + εt.
  - Missing values: account for about 0.2% of total sample size.
  - Stationarity checks: KPSS does not reject stationarity and ADF rejects the unit root in all cases.
  - Factor trend recovery: Φf = (Λ′Λ)−1(Λ′Φ) and f*t = ft + Φf,t (Equation 6).
  - Estimated autoregressive coefficient on the global factor = 0.6 (approximately 60% persistence month-to-month).
  - Rolling estimation: re-estimating trends and loadings over rolling windows of 72 observations produces an index within the 90% confidence intervals of the baseline.
  - Estimated factor trends revert after about 6 months in simulation of trend reversion.

### Baseline Trade Policy Activity (TPA) Index — main findings
- Definition: TPA Index is the normalized global factor (baseline value 0 for January 2010–December 2011); positive values indicate heightened activity relative to the baseline.
- Key temporal features:
  - Distinct peaks coincide with: the rise in trade tensions in 2025; the war in Ukraine (2022); the COVID-19 pandemic (2020); the U.S.–China tariff increases (2018–2019).
  - Additional peaks in 2013 (WTO Trade Facilitation Agreement conclusion) and 2016 (possible responses to China’s 2015 currency devaluation, surge in trade remedy investigations, Brexit uncertainty).
  - Rising trend that accelerates from 2020 onward with the 90% confidence interval remaining above zero from 2022 onward.
- Relationship to other indices:
  - Cointegration: Engle-Granger two-step procedure rejects null of no cointegration at the 10% level between TPA and each compared index.
  - Cointegrating regression R2 with Fernández-Villaverde et al. (2024) = 0.51.
  - R2 values with Caldara et al. (2020), Caldara and Iacoviello (2022), and Ahir et al. (2022) are 0.14, 0.18, and 0.03, respectively.
- Interpretation: index captures both short-term spikes linked to events and a secular rise tied to expanded use of trade policy for trade and non-trade objectives and geoeconomic fragmentation.

### Role of measure categories (facilitating, restrictive, other)
- Counterfactual approach: re-estimate global factor using baseline loadings for one category at a time; normalizations:
  - Average value set to zero over January 2010–December 2011.
  - Cyclical component rescaled so standard deviation matches baseline global factor over January 2010–December 2019.
  - Divide by absolute average TPA over January 2010–December 2011 for comparability.
- Findings:
  - Both facilitating and other measures show broadly rising trajectories, but “other measures” increasingly shape the global factor, with divergence widening over 2022–2025.
  - Crisis episodes (COVID-19, war in Ukraine): facilitating and other measures respond in tandem (local factors near zero), with global factor capturing common response.
  - 2018–2019 and 2025 trade tensions: local factors diverge—other measures accelerate while facilitating measures decelerate relative to the common signal.
- Subcategory decomposition within “other measures”:
  - Distinction: “Restrictive Measures” (tariffs and non-tariff measures that directly restrict trade, following Deardorff (2014)) vs. “Remaining Other Measures” (largely subsidies and other distortions).
  - Findings:
    - Remaining other measures show a broad-based increase over the sample (consistent with rising industrial policy).
    - The 2025 spike is explained predominantly by restrictive measures (tariffs and export controls).
    - Facilitating measures show a distinct rise following WTO TFA entry into force in 2017.

### Dynamics across country groups and weighting
- Group-specific estimation: global factor estimated separately for G20 and non-G20 country groups.
- Findings:
  - Both G20 and non-G20 record positive trends; increase begins earlier for G20.
  - Non-G20 recorded larger spikes during the COVID-19 pandemic.
  - 2017–2018 and 2025 peaks are more pronounced for G20 economies.
  - Within-group decomposition:
    - Remaining other measures (largely subsidies) increase markedly in both G20 and non-G20.
    - Among G20, restrictive measures diverge sharply in 2018–2019 and especially 2025.
    - Among non-G20, remaining other measures spike during COVID-19; 2025 increase in restrictive measures is less pronounced; facilitating measures subdued from 2022 onward.
  - Caveat: patterns could be partially affected by differences in coverage of underlying data across country groups.
- GDP-weighted index:
  - Construction uses countries’ pre-sample average share in global GDP (2005–2009) as time-invariant weights.
  - When weighted by economic size, rise in activity is less pronounced around COVID-19; peaks in 2018–2019 and early 2025 are more prominent.
  - Excluding U.S. and China confirms their important contribution, especially in 2025, but also shows broad-based acceleration in restrictive measures since 2022 (i.e., rise in protectionism not limited to the two largest economies).

### Robustness checks and sensitivity
- Model specification: alternative specification with a single global factor and no block restriction yields similar dynamics though higher values post-2019.
- Trend extraction: tested different weight functions and trend-filtering method of Tibshirani (2014); global factor shape robust to method.
- Rolling-window estimation: 72-month rolling window produces index within 90% confidence intervals of baseline; supports periodic updating.
- Input variables: tests adding number of implementing countries; counts-only vs counts+product coverage; inclusion of broader GTA measures (capital controls, FDI restrictions). Overall dynamics consistent; peak magnitudes vary with input dimension.
- Sample period: re-estimation with alternative end dates (October 2020 and October 2022) yields consistent loadings and dynamics.
- Missingness: estimation approach accommodates missingness (~0.2%).
- Additional notes: a version holding trends and loadings constant at baseline produces nearly identical historical estimates; divergence at sample end reflects informational gains from re-estimation.

### Conclusion and applications
- The TPA index provides a parsimonious monthly measure of global trade policy dynamics capturing event-driven spikes and a rising post-2020 trend.
- It distinguishes facilitating measures from restrictive/distortionary measures and documents episodes of co-movement and divergence across categories and country groups.
- Potential empirical uses:
  - Serve as a control variable to address omitted variable bias in studies of specific policy changes.
  - Use monthly frequency for event studies and local projections to study dynamic effects of trade policy shocks.
  - Examine how aggregate policy environment moderates international transmission of shocks (trade, investment reallocation).
  - Revisit determinants of trade policy and its cyclicality.
  - Operational use: robust to rolling-window estimation and suitable for periodic updating and nowcasting; Appendix demonstrates extension through January 2026 under alternative assumptions.

*Source: wpiea2025220-source-pdf - 2025. For a version through January 2026, based on additional assumptions, see Section A1.6 in Appendix A1.*

### References

### References

### Key datasets and methodological resources
- Primary data sources used:
  - WTO Trade Monitoring Database (TMDB), created in October 2008, tracks measures implemented by WTO Members and Observers through formal WTO channels and is released on a biannual basis.
  - Global Trade Alert (GTA), created in 2009, compiles announced and implemented measures from a variety of publicly available sources and provides near-real-time updates.
- Important methodological and econometric references cited include works on factor models, nowcasting, trend filtering, stationarity and unit-root testing, and measures of trade restrictiveness and geopolitical risk (selected examples: Bańbura and Modugno 2014; Forni et al. 2000; Tibshirani 2014; Kwiatkowski et al. 1992; Engle and Granger 1987).

### TMDB and GTA — overview and key differences
- TMDB:
  - Tracks only implemented measures that are officially communicated or notified by Members to the WTO, or identified by the WTO Secretariat from public sources.
  - Focuses on national measures with economy-wide effects.
  - Verification by WTO Members in terms of coverage, dates and content; roughly 90 to 95% of all TMDB measures have been confirmed (Pedersen and Diakantoni, 2020).
  - Updates: biannually; reporting period covers the 12 months leading up to mid-October.
- GTA:
  - Captures announced and implemented measures at national and sub-national levels, including firm-specific interventions.
  - Broader coverage (e.g., captures practical implications such as products potentially affected beyond legal stipulations).
  - Incorporates unofficial news outlets and may document every revision or modification of trade interventions (risk of overcounting).
  - No predefined reporting period; provides continuous updates as measures are discovered.
- Four notable dimensions of difference:
  - Scope (GTA broader, includes subnational and firm-specific measures; TMDB focuses on national, economy-wide measures).
  - Source and verification (TMDB relies on official government sources and verification; GTA trades verification depth for breadth).
  - Implementation status (TMDB records implemented measures only; GTA includes announcements).
  - Update frequency/discovery (TMDB biannual; GTA continuous with varying discovery lags).

### Data processing: discovery window and adjustments
- To correct for GTA’s retroactive additions and discovery lags, a consistent 12-month discovery window is implemented following Evenett and Fritz (2018):
  - For each measure, compare announcement date to discovery date (publication date in GTA) and retain the measure for a year only if it was discovered within 12 months.
  - The 12-month window runs from October to October to align with TMDB updates.
- Empirical justification and statistics:
  - On average, close to 80% of GTA measures are discovered within this 12-month window.
  - Figure A1.2 reports the share of GTA measures discovered within 12 months by announcement year with overall Mean = 77.9%.
  - G20 mean = 77.3%; non-G20 mean = 80.0%.
  - The GTA-only and TMDB-only trend decomposition preserves the upward trajectory of the trend, indicating robustness to GTA discovery practices (Figure A1.3).
- Extending the index beyond the baseline:
  - Baseline update covers through October 2025; the October-to-October 12-month discovery window fully elapsed for all months in the sample.
  - Extensions to January 2026 use a rolling-window approach with a bandwidth of 0.75 corresponding to approximately 144 observations on the current sample; an alternative rolling approach used 190-observation windows when adding months one at a time while dropping the oldest.
  - For months with incomplete discovery (November 2025–January 2026), month-specific adjustment factors can be applied assuming stable discovery patterns across time and country groups.
  - Figure A1.4 compares: baseline (through October 2025), rolling-window extension through January 2026 (red), and extension holding trends/loadings fixed at October 2025 baseline values (blue); the two extensions are broadly consistent with differences driven by re-estimation of the nonparametric trend.

### Categorization and classification of measures
- High-level classification:
  - “Facilitating” measures: those that enable trade, reduce bias against foreign interests, or increase transparency — labeled “facilitating” in TMDB and “green” in GTA. In the analysis, TMDB ‘facilitating’ and GTA ‘green’ are both classified as “facilitating”.
  - “Other” measures: all remaining measures not classified as facilitating (includes restrictive measures and other remaining measures).
- Identification of “restrictive” measures (following Deardorff 2014 and UN MAST classification):
  - Restrictive measures cover tariffs and non-tariff measures under MAST chapters E, C, I and selected subcomponents of P (excluding tax-based export incentives, export subsidies, trade finance, other export incentives, and export-related non-tariff measures, nes).
  - Measures in the “Other” group that do not belong to those MAST chapters are labeled “Other remaining” (mainly composed of subsidies and similar measures).
- When TMDB or GTA label a measure as facilitating/green, it is treated as facilitating regardless of MAST chapter.

### Dynamic factor model inputs and index construction
- Time-series variables used in the dynamic factor model (each defined by unit, category, and source):
  - GTA series:
    - Total number of measures — Facilitating
    - Total number of measures — Other
    - Average number of products — Facilitating
    - Average number of products — Other
  - TMDB series:
    - Total number of measures — Facilitating
    - Total number of measures — Other
    - Average number of products — Facilitating
    - Average number of products — Other
- Block structure and factors:
  - Baseline model uses one global factor and two local factors (Facilitating and Other) in the block structure.
  - Alternative specifications considered: global factor without blocks (local loadings set to zero), inclusion/exclusion of product-count series, and use of number-of-implementing-countries indicators.
- Normalization and presentation:
  - Factors are normalized relative to the January 2010–December 2011 reference period (baseline = 0).
  - Final reported factor is smoothed using a three-month moving average for clearer presentation (Figure A2.1 shows filtered vs unfiltered).
- Estimated loadings and relative contributions:
  - GTA series exhibit larger loadings on the global factor relative to TMDB series, reflecting greater variation in GTA inputs (Figure A2.2).
  - Decomposition of the baseline trend into GTA-only and TMDB-only contributions is exactly additive; GTA-only and TMDB-only trends sum to the baseline trend at every point while sharing the same cyclical component (Figure A1.3).

### Robustness checks and alternative specifications (selected findings)
- Trend extraction and nonlinearities:
  - Baseline trend extraction uses a weighted quadratic regression (local quadratic all).
  - Alternative trend extraction methods produce qualitatively similar trends: local quadratic with 33% weights, quadratic least-squares, and Tibshirani (2014) trend filter (Figure A2.5).
  - The time-varying slope of the TPA trend is shown and compared to linear and quadratic trend slopes, highlighting nonlinearities captured by the nonparametric approach (Figure A2.6).
- Sample and input-scope variations:
  - Alternative input sets tested include: adding the number of implementing countries per measure and using number-of-measures-only (excluding product counts). Results remain broadly consistent with baseline, though scale and short-term fluctuations vary (Figure A2.7).
  - Alternative scope including all GTA-recorded measures (including CAP, F, FDI, G, MIG, and unknown instruments) was tested; baseline result compared with facilitating-only and other-only factors (Figure A2.8).
  - Alternative time-period cutoffs were tested (e.g., stopping at Oct 2020 and Oct 2022) and produced consistent historical estimates (Figure A2.9).
- Subsample analysis:
  - TPA Index decomposed for G20 vs non-G20 economies, and by policy category (Facilitating, Restrictive, Remaining other) — each series normalized to the January 2010–December 2011 baseline (Figure A2.10).

*Supplemental Appendix and references as provided in the source PDF.*

### 2011. Its cyclical component is then rescaled to match the standard deviation of the baseline global factor over the lon

### Figures A2.12–A2.14 and Accompanying Notes

### Alternative Approach Accounting for Economic Size by Policy Category (Figure A2.12)
- Baseline depicted by the gray line.
- Panel A:
  - Baseline global factor estimated using a weighted sum of measures adopted globally of a given type as input indicators (rather than a simple sum).
  - Countries’ shares in the global GDP (an average during 2005-2009) are used as weights.
- Panel B:
  - Shows the global factor that loads exclusively on a given category of measures (facilitating, restrictive, or remaining other) using factor loadings from the baseline regression.
- Definitions:
  - Facilitating measures: any actions that clearly reduce restrictiveness or other distortions or otherwise ease trade (e.g., more efficient border procedures), including rollbacks of prior restrictions.
  - Restrictive measures: those that, following Deardorff (2014), directly restrict trade – tariffs and non-tariff measures (see Appendix A1 for more detail).
- Normalization and scaling procedure applied to each factor:
  - Each factor is normalized so that its average equals zero over January 2010–December 2011.
  - Its cyclical component is then rescaled to match the standard deviation of the baseline global factor over the longer window of January 2010–December 2019, which provides a more reliable estimate of each variant’s scale.
  - Finally, all factors are divided by the absolute value of the baseline global factor’s average over January 2010–December 2011, so that changes are expressed relative to this baseline period.

### The TPA Index Outside of U.S. and China (Figure A2.13)
- The figure shows the global factor that loads exclusively on a given category of measures (facilitating, restrictive, or remaining other) using factor loadings from the baseline regression.
- Dashed lines represent the global factor estimates on the sample of countries excluding the United States and China.
- Normalization and scaling procedure (applied as in Figure A2.12):
  - Each factor is normalized so that its average equals zero over January 2010–December 2011.
  - Its cyclical component is then rescaled to match the standard deviation of the baseline global factor over the longer window of January 2010–December 2019, which provides a more reliable estimate of each variant’s scale.
  - Finally, all factors are divided by the absolute value of the baseline global factor’s average over January 2010–December 2011, so that changes are expressed relative to this baseline period.

### Rolling-window Estimation of the Global Factor (Figure A2.14)
- Baseline result depicted by the black line.
- Rolling estimation:
  - Global factor with trend and loadings re-estimated using rolling samples of 72 observations (red line).
  - The trend is estimated using a quadratic function over each window.
- Uncertainty:
  - The factor estimated over rolling windows is within the 90% confidence intervals of the baseline, shown as shaded regions, and constructed using bootstrap resampling methods.

### Key Statistics and Windows (preserved exactly)
- Average used for GDP weights: 2005-2009.
- Normalization window for factor averages: January 2010–December 2011.
- Rescaling window for standard deviation: January 2010–December 2019.
- Rolling-window sample size: 72 observations.
- Confidence interval reported: 90% confidence intervals.

### References Cited in the Notes
- Deardorff, Alan V., Terms of Trade: Glossary of International Economics, 2 ed., World Scientific, 2014.
- Evenett, Simon and Johannes Fritz, “Working with a Growing Data Set: Technical Note on the Implications for Proper Reporting Lag Adjustment,” Technical Note, Global Trade Alert December 2018.
- Pedersen, Peter and Antonia Diakantoni, “Lessons Learned and Challenges Ahead for the WTO Trade Monitoring Exercise,” Staff Working Paper ERSD-2020-03, World Trade Organization February 2020.
- Tibshirani, Ryan J., “Adaptive piecewise polynomial estimation via trend filtering,” The Annals of Statistics, 2014, 42 (1), 285 – 323.
- United Nations Conference on Trade and Development, “International Classification of Non-Tariff Measures: 2019 Version,” UNCTAD Publication UNCTAD/DITC/TAB/2019/5, United Nations Conference on Trade and Development, Geneva 2019.

*Source: wpiea2025220-source-pdf - excerpt containing Figures A2.12–A2.14 and accompanying notes.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2025/english/wpiea2025220-source-pdf.pdf_