## _wp0845

## Source details

**Canonical URL:** [_wp0845](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0845.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0845.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0845.pdf.json)

---

### Introduction — purpose and overview
- Paper's aim: provide a systematic empirical examination of the role played by tradables and nontradables in aggregate investment.
- Main empirical summaries:
  - "on average around 60 percent of aggregate investment expenditures are spent on nontradables."
  - Average expenditure share on nontradables varied between 0.54-0.62 over the 1960-2004 period.
  - Aggregate investment expenditure shares on tradable and nontradable goods show no correlation with income and are very similar across regions (Africa, South-East Asia, Europe, Latin America).
  - Between some sample countries expenditure shares do exhibit sizable differences.
  - If residential structures are excluded from investment data the weight of nontradables decreases to 46 percent of aggregate investment.
  - 80-90 percent of aggregate investment expenditures are spent on output from two sectors: equipment from manufacturing (tradable) and structures from construction (nontradable).
- Relevance: results apply to small open economy models distinguishing tradables/nontradables and to closed economy models differentiating equipment and structures.

### Empirical findings from input-output tables and sector mapping
- Data coverage and prominent sectoral shares:
  - Input-output table data examined for different countries and years between 1968 and 2001; U.S. 1997 benchmark emphasized.
  - Around 90 percent of all investment expenditures are spent on the output of only two sectors: manufacturing and construction.
  - Retail/wholesale trade and real estate/business services are the only two other sectors with significant fractions; together the four sectors account for 98 percent of aggregate investment expenditures.
  - Weight of manufacturing and construction gradually decreased from 0.9 in the 1970s to 0.8 by year 2000.
  - For the U.S. in 1997, the same four sectors comprised 97.4 percent of aggregate investment expenditures; real estate/business services accounted for 12.1 percent after adding software and related services.
  - Retail/wholesale trade services account for 5-7 percent of investment expenditures throughout 1968-2001.
- Mapping sectors to tradable vs nontradable:
  - Distribution sector (retail/wholesale trade) assumed nontradable.
  - Tradability measured as (exports + imports) / gross output.
  - Classification outcomes (27 countries, 1995-2001 IO tables): Construction classified nontradable; Manufacturing classified tradable; Real estate/business services mixed — for the U.S. 1997 subsector split implies the sector is treated as 2/3 nontradable and 1/3 tradable (tradable subsectors account for 31 percent of that sector's investment expenditures).
  - Sensitivity: because nontradable construction and tradable manufacturing dominate investment, classification of remaining sectors has limited effect.

### National accounts (GFCF) evidence, coverage, and measurement issues
- Data sources and scope:
  - UN detailed NA (SNA68) covers 1950-1970; OECD detailed NA (SNA93) covers 1970-2004; Penn World Table (PWT) benchmarks for selected years provide broader cross-section (up to 115 countries).
- Measurement caveats:
  - GFCF measures expenditures at purchaser's prices and does not separately capture expenditures on the retail/wholesale trade sector; most expenditures on that sector are recorded together with equipment/machinery.
  - Consequently, GFCF data should overestimate the weight of tradables by 0.01-0.10.
  - Switch from SNA68 to SNA93 moves expenditures on computer related services (e.g., software) from intermediate consumption into investment and affects post-1990 comparability.
- Comparison of IO and national accounts:
  - National accounts data overestimate the share of tradables in investment by up to 0.09.
  - Despite this bias, national accounts data offer wider coverage (yearly data starting from 1950 and cross-sections of up to 115 countries) and better comparability; paper builds on GFCF evidence.

### Time-series evidence (OECD and UN national accounts)
- OECD and UN findings:
  - Six largest OECD economies — investment expenditures on nontradable goods for any given year are in 0.47-0.68 range.
  - No systematic trend over available periods for most countries; persistent cross-country differences exist (examples: France, UK).
  - OECD sample (Table 7): all country-year observations between 0.35-0.77; sample average 0.59; highest average Iceland 0.67; lowest Slovak Republic 0.42.
  - Persistence across three OECD periods: correlations 0.63, 0.46, and 0.81 for pairwise decade comparisons.
  - Time trend regressions (countries with ≥30 observations): for eight out of thirteen countries the time trend is not significantly different from zero at the 5 percent confidence level; point estimates (except Denmark) between -0.024 to 0.027 per decade.
  - UN dataset: at least one observation for 113 countries; cross-country averages range from 0.34 (Saint Kitts and Nevis) to 0.97 (Kyrgyzstan); decadal persistence correlations in 0.64-0.86 range.
  - Pooled/panel regressions (Table 10):
    - OECD national accounts: pooled time trend -0.003 per decade (not significant).
    - UN data pooled: time trends per decade vary between -0.014* to -0.020* depending on sample/specification; panel with country dummies shows trends -0.017*, -0.012*, -0.020* (all * significant at 5%).
  - Aggregate averages: no economically significant time trends; OECD and UN averages very similar.

### Cross-section and regional evidence
- Correlation with income:
  - UN cross-sections (1950-1997): correlation between expenditure share and PPP adjusted income per capita typically within 0.00-0.30; average correlation during 1950-97 is 0.10.
  - In most years a zero correlation cannot be rejected at 5% confidence level; only four out of 48 years reject zero correlation.
  - Illustration: fitted trend implies a country with per capita income of 10 percent of the average OECD level exhibits an expenditure share on nontradables which is 0.01-0.05 lower than in OECD countries.
  - PWT benchmark cross-sections: positive correlations in six sample years; in five out of six years correlation between 0.04 and 0.31.
- Regional patterns:
  - UN region grouping (Africa, Europe, Latin America, South East Asia): regional averages do not deviate from sample mean by more than 0.05; Europe average higher than Africa consistent with small positive income correlation.
  - PWT 1996 seven-region grouping: coefficients range between 0.51 and 0.59, with exception Africa at 0.23 (PWT 1996); Africa averages in earlier PWT benchmarks show no deviation (0.54, 0.57).

### Effect of excluding residential structures
- Excluding residential structures lowers nontradable share:
  - OECD data: average decreases from 0.60 to 0.47 when residential construction is excluded.
  - UN data: decrease from 0.58 to 0.45 in UN data.
- Empirical note: expenditures on residential structures account for a quarter of all investment expenditures and do not vary systematically with the level of income (correlation coefficients for 1960-1996 in -0.25 to 0.07 range, average -0.09).

### Key numeric highlights (preserved exactly)
- Around 60 percent of aggregate investment expenditures are spent on nontradables.
- Average nontradable share varied between 0.54-0.62 over the 1960-2004 period.
- Nontradable weight falls to 46 percent if residential construction is excluded.
- 80-90 percent of aggregate investment expenditures are on equipment (manufacturing) and structures (construction).
- Manufacturing and construction weight fell from 0.9 in 70s to 0.8 by year 2000.
- For the U.S. in 1997, four sectors comprised 97.4 percent of aggregate investment expenditures; real estate/business services = 12.1 percent.
- Retail/wholesale trade services account for 5-7 percent of investment expenditures (1968-2001).
- Depending on period, 59-64 percent of aggregate investment expenditures are on nontradables.
- For the U.S., tradable subsectors within real estate/business services account for 31 percent of that sector's investment expenditures; sector treated as 2/3 nontradable and 1/3 tradable.
- Input-output table coverage analyzed for years 1968-2001; U.S. benchmark 1997 emphasized.

### Modeling implications, applications, and quantitative experiments
- Aggregate accounting and formal relations:
  - Paper formalizes an aggregate resource constraint with tradable (T) and nontradable (N) consumption, investment, and output shares.
  - Assumes constant fraction γ of investment expenditures is spent on tradables (and γ−1 on nontradables).
  - Equipment/machinery investment share denoted τ and does not vary systematically with income.
  - Aggregate investment rate expressed as γ+τ and empirically approximately constant across countries and time.
- Modeling recommendations:
  - To generate constant investment expenditure shares, impose a Cobb-Douglas (unitary elasticity of substitution) aggregation for the two investment goods.
  - Empirical evidence does not support modeling assumptions that only tradables (or only nontradables) are transformed into investment goods, nor that investment uses tradables and nontradables in the same proportion as consumption.
  - Allowing different depreciation rates by capital type requires adjusting capital-income share relations (θ relates to γ via θ = [ (r+δT) γ + (r+δN) (1−γ) ] / (r+δT)).
  - Models with capital-skill complementarity and CES nests may be incompatible with observed constancy of investment shares unless restrictive parameter values (e.g., zero elasticities) are imposed.
- Two-sector growth model application and quantitative conclusions:
  - Calibration sets investment rate to 0.23 (average for high-income reference group); parameters include γ, θ, μ.
  - Extreme parametrization (μ = 0, γ = 1: tradable investment and nontradable consumption) generates variation in international-price investment rates that exceeds observed data variation.
  - Empirically motivated parametrization (γ = 0.4; θ within reported range 0.144 <= θ; μ set so consumption tradable share in high-income country is 0.25):
    - As relative productivity falls, model investment rates initially decrease slightly because investment tradable share (0.44) > consumption tradable share (0.25).
    - With less than unitary consumption elasticity, consumption tradable share increases as income falls; eventually consumption tradable share exceeds investment tradable share and investment rates increase for low incomes.
    - Therefore the model cannot account for the observed monotonic decrease of investment rates with income.
  - Upper-bound experiment (γ = 0.4 and μ = 0) — model can account for about half of observed differences in investment rates between rich and poor countries.
  - Quantitative conclusion: with empirically documented prominent role of nontradables in investment (γ ≈ 0.4), differences in relative sectoral productivities and relative prices can account for only a small fraction (at most about half under extreme consumption assumptions) of observed differences in international-price investment rates across countries; in some calibrations decreasing relative tradable productivity can raise, not lower, measured investment rates in poor countries.
- Additional applications (Section 6):
  - Exact mix of tradables and nontradables in investment and aggregate capital stock affects speed of transitional dynamics in two-sector open economy growth models; larger nontradable share slows transition.
  - Construction may play a special role as collateral; empirical result that residential structures account for a quarter of all investment expenditures informs models with collateral constraints.

### Appendix I — IO vs national accounts numeric comparisons (selected)
- IO vs national accounts differences for investment expenditures on manufacturing output (selected entries, preserving exact values as in source):
  - Australia (1970-1990 sequence): IO 0.310, 0.273, 0.321, 0.273; NA 0.470, 0.435, 0.495, 0.452; differences 0.160, 0.162, 0.174, 0.179.
  - Denmark (series): IO 0.259, 0.331, 0.329, 0.410, 0.399; NA 0.306, 0.370, 0.372, 0.468, 0.460; differences 0.047, 0.039, 0.043, 0.058, 0.061.
  - Germany (series): IO 0.423, 0.471, 0.486, 0.500; NA 0.399, 0.431, 0.445, 0.462; differences -0.024, -0.040, -0.041, -0.038.
  - Japan (series): IO 0.360, 0.276, 0.265, 0.338, 0.311; NA 0.433, 0.342, 0.326, 0.402, 0.398; differences 0.073, 0.066, 0.061, 0.064, 0.087.
  - USA, 1997: IO 0.354; NA 0.389; difference 0.035.
  - Selected 1995-2001 comparisons include Hungary 2000 IO 0.378 NA 0.466 difference 0.088; Australia 1994/95 IO 0.274 NA 0.361 difference 0.088; UK 1998 IO 0.422 NA 0.499 difference 0.077.
- Conclusion: for majority of countries expenditure share on manufacturing output is higher in national accounts data; differences generally consistent with expenditure share on distribution sector (bias in 0.003-0.088 range); some country-specific incompatibilities exist.

### Robust empirical regularities and reconciliation with other studies
- Regularities:
  - Investment expenditure shares on tradables and nontradables have been close to constant over the last 50 years and show at most a small positive correlation with income.
  - Nontradables dominate both consumption and investment expenditures: roughly 60 percent of investment expenditures on nontradables and 40 percent on tradables.
  - Relative price of nontradables in investment strongly positively correlated with income; relative price has doubled on average in largest OECD economies over last 30 years and more than tripled in the U.S. over same period.
  - Despite large variation in relative prices, investment shares remain nearly constant.
- Reconciliation:
  - Burstein et al. (2004) reported correlation -0.69 between construction share and income using 19 IO observations; replication with national accounts for same set yields -0.49 here — differences largely due to choice of observations and outlier years.

### Conclusions and modeling recommendation
- Empirical conclusions:
  - Investment expenditure shares on tradables and nontradables are nearly constant across countries and over time (1960-2004), with average nontradable share around 60 percent.
  - Excluding residential structures reduces nontradable share materially (to 46 percent in OECD data).
- Modeling recommendation:
  - A realistic two-sector modeling of capital accumulation can be achieved by assuming a Cobb-Douglas (unitary elasticity of substitution) aggregation for the two investment components, which preserves empirical constant investment shares while allowing sectoral production flexibility.
  - Modeling shortcuts assuming investment is tradable-only or mirrors consumption composition are not empirically supported and can materially affect quantitative outcomes in cross-country investment analyses.

*Content derived from the provided IMF working paper excerpt.*

### References..............................................................................................................

### _wp0845 - References

### Introduction — purpose and overview
- Paper's aim: provide a systematic empirical examination of the role played by tradables and nontradables in aggregate investment.
- Main empirical summaries:
  - "on average around 60 percent of aggregate investment expenditures are spent on nontradables."
  - Average expenditure share on nontradables varied between 0.54-0.62 over the 1960-2004 period.
  - Aggregate investment expenditure shares on tradable and nontradable goods show no correlation with income and are very similar across regions (Africa, South-East Asia, Europe, Latin America).
  - Between some sample countries expenditure shares do exhibit sizable differences.
  - If residential structures are excluded from investment data the weight of nontradables decreases to 46 percent of aggregate investment.
  - 80-90 percent of aggregate investment expenditures are spent on output from two sectors: equipment from manufacturing (tradable) and structures from construction (nontradable).
- Relevance: results apply to small open economy models distinguishing tradables/nontradables and to closed economy models differentiating equipment and structures.

### Empirical findings on investment structure (input-output table evidence)
- Data coverage and sources:
  - Input-output table data examined for different countries and years between 1968 and 2001.
  - For post-1990 period focus on the 1997 benchmark input-output table for the U.S.
- Dominant sectors in investment expenditures:
  - Around 90 percent of all investment expenditures are spent on the output of only two sectors: manufacturing and construction.
  - Retail/wholesale trade and real estate/business services are the only two other sectors with significant fractions.
  - Together the four sectors account for 98 percent of aggregate investment expenditures.
- Time evolution and new components:
  - Weight of manufacturing and construction gradually decreased from 0.9 in the 1970s to 0.8 by year 2000.
  - Emergence of 'software' and 'computer system design and related services' as non-negligible components of investment expenditures in the 1990s.
  - For the U.S. in 1997, the same four sectors comprised 97.4 percent of aggregate investment expenditures; real estate/business services accounted for 12.1 percent after adding software and related services.
  - Retail/wholesale trade services account for 5-7 percent of investment expenditures throughout 1968-2001.
- Aggregate summary from input-output tables:
  - Depending on the period, 59-64 percent of aggregate investment expenditures are spent on nontradable goods.

### Mapping sectors to tradables vs nontradables
- Method:
  - Distribution sector (retail/wholesale trade) assumed nontradable.
  - Tradability measured as (exports + imports) / gross output.
- Sector classifications (27 countries, 1995-2001 input-output tables):
  - Construction: tradability uniformly lower than distribution sector → classified nontradable.
  - Manufacturing: tradability uniformly higher than distribution sector → classified tradable.
  - Real estate/business services: mixed tradability; U.S. disaggregation used:
    - Subsector classification in U.S. 1997 benchmark: real estate services (nontradable); computer system design and related services (nontradable); software (tradable); other business services (tradable).
    - For the U.S. the two tradable subsectors account for 31 percent of total investment expenditures on the sector's output; accordingly the sector is defined as 2/3 nontradable and 1/3 tradable for aggregation purposes.
- Sensitivity:
  - Alternative approach of defining real estate/business services fully nontradable was considered; results presented use the approach less favorable to the paper's main findings (i.e., the one that makes aggregate investment expenditure shares less constant across income and time).
  - Because nontradable construction and tradable manufacturing dominate investment, the classification of remaining sectors has limited effect.

### Alternative data sources and measurement issues
- Use of national accounts "GDP by expenditure" gross fixed capital formation (GFCF) data:
  - GFCF divides investment spending into residential structures, nonresidential structures and equipment/machinery, allowing separation between construction (nontradable) and manufacturing (tradable).
  - Under SNA93, GFCF also accounts separately for real estate/business services output.
- Measurement caveat:
  - GFCF measures expenditures at purchaser's prices and does not separately capture expenditures on the retail/wholesale trade sector; most expenditures on that sector are recorded together with equipment/machinery.
  - Consequently, GFCF data should overestimate the weight of tradables by 0.01-0.10 (range of investment expenditure shares on nontradable output of the retail/wholesale trade sector, per Tables 2-4).
- SNA definition change:
  - Switch from SNA68 to SNA93 moves expenditures on computer related services (e.g., software) from intermediate consumption into investment and is more explicit about investment expenditures on services; this affects post-1990 data comparability.

### Integration with known empirical regularities and modeling implications
- Price regularity:
  - Relative price of nontradable goods in terms of tradable goods exhibits a strong positive correlation with income in cross-section and time-series; investment price data conform to this regularity.
- Modeling implications:
  - Combined with large variation in relative prices, results suggest that the aggregate economy investment process can be modeled using a unitary elasticity of substitution between tradable and nontradable goods (Cobb-Douglas) plus a country-specific investment share parameter.
  - Empirical evidence does not support modeling assumptions that only tradables or only nontradables are transformed into investment goods, nor that the role of tradables and nontradables in investment equals their role in consumption.
  - For models that use detailed investment expenditure data and assume constant country-specific expenditure shares, the paper provides supporting empirical evidence.
- Example model application:
  - Two-sector growth model exercise: when standard literature assumptions are used (investment mostly tradable, consumption mostly nontradable), productivity differences between tradable and nontradable sectors can explain low investment rates in poor countries.
  - However, when model functional forms and parameters are restricted to comply with the paper's empirical results (investment expenditures are mostly nontradable), productivity differences can at best explain a small fraction of variation in investment rates.

### Key numeric highlights (preserved exactly as in source)
- Around 60 percent of aggregate investment expenditures are spent on nontradables.
- Average nontradable share varied between 0.54-0.62 over the 1960-2004 period.
- Nontradable weight falls to 46 percent if residential construction is excluded.
- 80-90 percent of aggregate investment expenditures are on equipment (manufacturing) and structures (construction).
- Manufacturing and construction weight fell from 0.9 in 70s to 0.8 by year 2000.
- For the U.S. in 1997, four sectors comprised 97.4 percent of aggregate investment expenditures; real estate/business services = 12.1 percent.
- Retail/wholesale trade services account for 5-7 percent of investment expenditures (1968-2001).
- Depending on period, 59-64 percent of aggregate investment expenditures are on nontradables.
- For the U.S., tradable subsectors within real estate/business services account for 31 percent of that sector's investment expenditures; sector treated as 2/3 nontradable and 1/3 tradable.
- Input-output table coverage analyzed for years 1968-2001; U.S. benchmark 1997 emphasized.

*Content derived from the provided IMF working paper excerpt.*

### Appendix I compares investment data from input-output tables and national accounts. In line with

### Appendix I compares investment data from input-output tables and national accounts

### Comparison of input-output tables and national accounts
- National accounts data overestimate the share of tradables in investment by up to 0.09.
- Despite this bias, national accounts data offer advantages:
  - Wider coverage: yearly data starting from 1950 and cross-sections of up to 115 countries.
  - Better comparability across time and space than input-output table data.
- The paper builds on time-series and cross-section evidence from GFCF data of national accounts.

### Datasets used
- United Nations (UN) detailed national accounts statistics:
  - Compiled using SNA68 definitions.
  - Only dataset covering the period between 1950 and 1970.
  - Discontinued in 1997 due to the switch to SNA93 definitions.
- OECD detailed national accounts statistics:
  - Compiled using SNA93 definitions.
  - Only dataset covering the period from 1997 onwards.
- Penn World Table (PWT) benchmark investment expenditure data:
  - Based on Summers et al. (1995) and Heston et al. (2002).
  - Complemented with data from Nehru and Dhareshwar (1993).
  - Not annual frequency but offers the largest cross-section sample with 115 countries.

### Time-series evidence (OECD and UN national accounts)
- Six largest OECD economies (Figure 1) — investment expenditures on nontradable goods for any given year are in 0.47-0.68 range.
- No systematic trend over available periods for most countries, with a possible exception of France.
- Persistent cross-country differences (example: France and UK).
- OECD sample (Table 7) summary statistics:
  - All country-year observations of nontradable investment expenditure share are between 0.35-0.77.
  - Sample average is 0.59.
  - Highest average expenditure share: Iceland, 0.67.
  - Lowest average expenditure share: Slovak Republic, 0.42.
- Persistence across three OECD periods (eleven-year periods):
  - Correlation between 1970-80 and 1981-91: 0.63.
  - Correlation between 1970-80 and 1992-2002: 0.46.
  - Correlation between 1981-91 and 1992-2002: 0.81.
- Time trend regressions for OECD countries (countries with ≥30 annual observations):
  - Time trends expressed as change in aggregate investment expenditure shares on nontradables over a decade.
  - For eight out of thirteen countries, the time trend is not significantly different from zero at the 5 percent confidence level.
  - With the exception of Denmark, point estimates of time trends are between -0.024 to 0.027 per decade.
- UN dataset (Table 8) highlights:
  - At least one observation for 113 countries.
  - Cross-country averages range from 0.34 (Saint Kitts and Nevis) to 0.97 (Kyrgyzstan).
  - Decadal persistence (Table 9) for periods 1950-59, 1960-69, 1970-79, 1980-89 and 1990-97:
    - Correlations between subsequent decades are in 0.64-0.86 range.
- Time trend regressions in UN data (countries with ≥30 annual observations):
  - Generally similar estimates to OECD results.
  - Several low income countries (e.g., Lesotho and Guatemala) have considerably larger point estimates of time trends than OECD countries.
- Pooled and panel regressions (Table 10) findings:
  - OECD national accounts: small and negative trend, negligible economically, not significantly different from zero at the 5 percent confidence level.
  - UN dataset: time trends somewhat larger and statistically significant at the 5 percent confidence level; point estimates vary between -0.020 to -0.012 per decade depending on sample and specification.
  - Time trends for non-OECD countries have higher standard errors; point estimates range from -0.020 to -0.016 per decade, similar to OECD ranges.
- Aggregate averages (Figure 2):
  - No economically significant time trends in average expenditure shares for OECD and UN datasets.
  - Average expenditure shares are very similar in the two datasets, indicating OECD and non-OECD countries spend a similar share of investment resources on nontradable goods.

### Cross-section evidence
- Correlation with income (UN dataset, Table 11):
  - Table 11 presents cross-section results for each year between 1950 and 1997.
  - Correlation between expenditure share and PPP adjusted income per capita:
    - In all but a few sample years the correlation is within 0.00-0.30 range.
    - Zero correlation rejected at 5% confidence level for four out of 48 sample years.
    - Average correlation during 1950-97 is 0.10.
  - Regression of nontradable investment share on log real per capita income:
    - For all but five years a zero trend cannot be rejected at 5% confidence level.
  - Overall: small and positive correlation between expenditure shares and per capita income, not significantly different from zero.
- Illustration (Figure 3):
  - Years plotted: 1960, 1970, 1980 and 1990.
  - A fitted linear trend suggests a country with per capita income of 10 percent of the average OECD level exhibits an expenditure share on nontradables which is 0.01-0.05 lower than in OECD countries.
- PWT benchmark cross-sections (Table 12):
  - For six sample years, correlation is positive.
  - In five out of six years the correlation is between 0.04 and 0.31.

### Regional patterns
- UN data by region (Figure 4) — four country groups: Africa, Europe, Latin America and South East Asia:
  - Regional investment shares are simple arithmetic averages.
  - Europe average expenditure share higher than Africa, consistent with small positive income correlation.
  - Expenditure shares in the four regions do not deviate from the sample mean by more than 0.05.
- PWT 1996 benchmark (Table 13):
  - Seven-region grouping shows little variation: coefficients range between 0.51 and 0.59.
  - Notable exception: Africa in the PWT 1996 benchmark dataset has a considerably lower share than other regions.
  - Africa averages for 1985 and 1980 PWT benchmarks show no deviation from sample averages.
- GDP-weighted observations:
  - Results are not affected when country observations are weighted by total GDP in international prices.
  - Average correlation between total GDP in international prices and expenditure shares in UN dataset is 0.07.
  - Correlations across years vary in -0.10 to 0.15 range.

### Effect of excluding residential structures
- Both OECD and UN national accounts generally distinguish investment on residential and other structures.
- Excluding residential structures lowers the average expenditure share on nontradables:
  - OECD data: average decreases from 0.60 to 0.47 when residential construction is excluded.

*Source: Appendix I, _wp0845 - Appendix I compares investment data from input-output tables and national accounts. In line with*

### 0.58 to 0.45 in UN data. None of the other findings of this section are significantly altered.

### _wp0845 - 0.58 to 0.45 in UN data. None of the other findings of this section are significantly altered.

### Empirical findings on investment composition and prices
- Expenditures on residential structures account for a quarter of all investment expenditures and do not vary systematically with the level of income.
- For the 1960-1996 period correlation coefficients are in -0.25 to 0.07 range with an average correlation of -0.09.
- Investment expenditure share on nontradables has been close to constant over the last 50 years and exhibits no significant correlation with the level of income.
- On average, expenditures on nontradable and tradable goods account for 60 and 40 percent of investment expenditures, respectively.
- Relative price behavior:
  - In poorest developing countries relative price of nontradables in investment is a third of the same price in advance economies (Figure 5 in source).
  - The relative price of nontradables in investment has on average doubled in the largest OECD economies over the last 30 years; in the U.S. it has more than tripled over the same period (Figure 6 in source).
- Despite large and systematic variation in relative prices across time and income levels, investment expenditure shares on tradables and nontradables remain nearly constant.

### Aggregate accounting and empirical regularities (summary of formal relations)
- Aggregate resource constraint (notation in source): tradable (T) and nontradable (N) consumption, investment, and output shares expressed as shares of aggregate output (equation (1) in source).
- Empirical summaries in source:
  - Constant fraction γ of investment expenditures is spent on tradables (and γ−1 on nontradables) — equation (2).
  - Equipment/machinery investment share denoted τ and does not vary systematically with income — equation (3).
  - Aggregate investment rate expressed as γ+τ (equation (4)) and empirically approximately constant across countries and time.
- Implication for consumption and output shares:
  - With constant investment shares, sectoral expenditure share changes for output and consumption have the same sign (equation (5)).
  - Combined with increasing relative price of nontradables with income, CES aggregation for consumption implies elasticity of substitution less than unitary (supported by literature cited in source).

### Implications for modeling choices
- Broad modeling implication: two-sector open-economy models must explicitly assume the role of tradable and nontradable goods in investment; empirical evidence supports substantial nontradable share in investment.
- Simple aggregate-capital specification:
  - To generate constant investment expenditure shares, impose a Cobb-Douglas aggregation for the two investment goods (equation (9) in source). Under this, investment expenditure shares γ and γ−1 are constant regardless of elasticity of substitution in production.
- Models with separate tradable and nontradable capital:
  - Allowing different depreciation rates by capital type requires restricting the production function to a specific form (equation (12)) where θ (relative capital income shares) relates to γ via equation (13): θ = [ (r+δT) γ + (r+δN) (1−γ) ] / (r+δT).
  - If δN = δT then θ = γ and models coincide; if δN ≠ δT then relative capital income shares differ from investment expenditure shares by the depreciation adjustment.
- Models with capital-skill complementarity:
  - If production nests skilled and unskilled labor and capital as in equation (15) (double-nested CES), trends in relative prices imply incompatibility with empirical constancy of investment shares unless ρ = 0 and σ = 0 (i.e., zero elasticities in those nests). CES with non-unitary elasticities and trending relative prices leads to trending expenditure shares and eventual nonuse of some factors.

### Reconciliation with other empirical studies
- Burstein et al. (2004) reported a correlation of -0.69 between investment share on construction and per capita income using 19 country-year IO observations dated over 1990-1999.
- Replication using national accounts data for the same country-year set yields correlation coefficient -0.49 in this paper. Much of the difference is attributable to the particular 19 country-year observations used by Burstein et al. (2004), with 9 of 19 observations for 1995-96 which are outlier years in the broader sample.

### Application to the growth literature: two-sector growth model results
- Model setup highlights:
  - Representative consumer with CES consumption aggregator and Cobb-Douglas production in each sector (equations (16), (17), (18) in source).
  - Investments aggregated using Cobb-Douglas with share γ as empirically motivated (equation (9) in source).
- Key analytical results:
  - Domestic-price investment rate in the model is δ + ... (expression in source, equation (19)) and does not depend on productivities; thus domestic-price investment rates are the same across model economies (supporting data).
  - Relative price of nontradables equals ratio of sectoral productivities, pN = AN/AT (Balassa–Samuelson effect); lower tradable productivity reduces relative price of nontradables and lowers measured international-price investment rates when investment is tradable-intensive relative to consumption.
- Quantitative experiments and calibration:
  - Calibration sets investment rate to 0.23 (average for high-income reference group). Parameters calibrated include γ, θ, and μ (see source).
  - Extreme parametrization: tradable investment and nontradable consumption (μ = 0, γ = 1) — model generates variation in international-price investment rates that exceeds observed data variation (dotted line in Figure 7 in source).
  - Empirically motivated parametrization: γ = 0.4; θ set within reported range 0.144 <= θ (Stockman and Tesar (1995)); μ set so consumption tradable share in high-income country is 0.25. Under this calibration:
    - As relative productivity falls, model investment rates initially decrease slightly because investment tradable share (0.44) > consumption tradable share (0.25).
    - With less than unitary consumption elasticity, consumption tradable share increases as income falls; eventually consumption tradable share exceeds investment tradable share and investment rates increase for low incomes.
    - Thus, with empirically motivated parametrization, the model cannot account for the observed monotonic decrease of investment rates with income.
  - Upper-bound experiment: γ = 0.4 and μ = 0 (only nontradables in consumption) — model can account for about half of observed differences in investment rates between rich and poor countries.
- Quantitative conclusion:
  - With reasonable parameter values and the empirically documented prominent role of nontradables in investment (γ ≈ 0.4), differences in relative sectoral productivities and relative prices can account for only a small fraction (at most about half under extreme consumption assumptions) of observed differences in international-price investment rates across countries. In some calibrations decreasing relative tradable productivity can raise, not lower, measured investment rates in poor countries.

### Conclusions and modeling recommendation
- Empirical regularities:
  - Investment expenditure shares on tradables and nontradables have been close to constant over the last 50 years and show at most a small positive correlation with income.
  - Nontradables dominate both consumption and investment expenditures: roughly 60 percent of investment expenditures on nontradables and 40 percent on tradables (source statement).
- Modeling implication:
  - A realistic two-sector modeling of capital accumulation can be achieved with relatively little added complexity by assuming a unitary elasticity of substitution between the two investment components (i.e., Cobb-Douglas aggregation of investment components).
  - This assumption preserves empirical constant investment shares while allowing flexibility in production and factor intensities across sectors.
- Broader implication:
  - Common modeling shortcuts—assuming only tradables can be invested in or assuming investment uses tradables in the same proportion as consumption—are not empirically supported and can materially affect quantitative model outcomes, especially for cross-country comparisons of investment rates.

*Source: _wp0845 - 0.58 to 0.45 in UN data. None of the other findings of this section are significantly altered.*

### Section 6. Here we point out two additional applications. First, the exact mix of tradables and

### Section 6

### Additional applications

- The exact mix of tradables and nontradables in investment and aggregate capital stock affects the speed of transitional dynamics in a two sector open economy growth model, with a larger share for nontradables slowing down the transition process.
- This channel can be used as an alternative to assuming excessive frictions for production factors and/or prices to obtain more plausible transitional dynamics.
- Construction, beyond being nontradable, may play a special role because real estate is the quintessential collateral asset.
- Empirical results show that residential structures account for quarter of all investment expenditures; these results can be applied to models that include collateral constraints in both cross-country and time-series settings.

### Appendix I — Compatibility of national accounts and input-output table data

- The appendix compares investment expenditure shares on the output of the manufacturing sector between input-output tables (basic prices) and national accounts (purchaser's prices). The difference stems from the added value of the distribution sector.
- Table A1 (pre-1990 input-output tables) findings:
  - For 7 out of 9 countries expenditure shares are slightly higher in national accounts data, with the bias in 0.003-0.087 range.
  - For Germany and Australia differences are larger, suggesting incompatibility for those cases.
- Table A2 (1995-2001 OECD/latest input-output tables) findings:
  - For the majority of countries expenditure share is higher in national accounts data.
  - The difference does not exceed 0.088, consistent with the expenditure share on the output of the distribution sector.
- Selected numeric comparisons from Table A1 (investment expenditures on manufacturing output, input-output tables vs national accounts, 1970-1990):
  - Australia: input-output tables 0.310, 0.273, 0.321, 0.273; national accounts 0.470, 0.435, 0.495, 0.452; differences 0.160, 0.162, 0.174, 0.179.
  - Denmark: input-output tables 0.259, 0.331, 0.329, 0.410, 0.399; national accounts 0.306, 0.370, 0.372, 0.468, 0.460; differences 0.047, 0.039, 0.043, 0.058, 0.061.
  - France: input-output tables 0.322, 0.321; national accounts 0.381, 0.377; differences 0.059, 0.056.
  - Germany: input-output tables 0.423, 0.471, 0.486, 0.500; national accounts 0.399, 0.431, 0.445, 0.462; differences -0.024, -0.040, -0.041, -0.038.
  - Italy: input-output tables 0.393; national accounts 0.466; difference 0.073.
  - Japan: input-output tables 0.360, 0.276, 0.265, 0.338, 0.311; national accounts 0.433, 0.342, 0.326, 0.402, 0.398; differences 0.073, 0.066, 0.061, 0.064, 0.087.
  - Netherlands: input-output tables 0.319, 0.342, 0.324, 0.422; national accounts 0.381, 0.406, 0.365, 0.490; differences 0.062, 0.064, 0.041, 0.068.
  - UK: input-output tables 0.427, 0.472, 0.437, 0.400; national accounts 0.454, 0.494, 0.470, 0.437; differences 0.027, 0.022, 0.033, 0.037.
  - USA: input-output tables 0.366, 0.392, 0.373, 0.388, 0.387; national accounts 0.369, 0.413, 0.426, 0.425, 0.444; differences 0.003, 0.021, 0.053, 0.037, 0.057.
- Selected numeric comparisons from Table A2 (input-output tables vs national accounts, 1995-2001):
  - USA, 1997: input-output tables 0.354; national accounts 0.389; difference 0.035.
  - France, 2000: input-output tables 0.339; national accounts 0.326; difference -0.013.
  - Germany, 2000: input-output tables 0.420; national accounts 0.399; difference -0.020.
  - Austria, 2000: input-output tables 0.369; national accounts 0.406; difference 0.037.
  - Denmark, 2000: input-output tables 0.336; national accounts 0.403; difference 0.067.
  - Finland, 2000: input-output tables 0.315; national accounts 0.303; difference -0.012.
  - Hungary, 2000: input-output tables 0.378; national accounts 0.466; difference 0.088.
  - Ireland, 1998: input-output tables 0.279; national accounts 0.351; difference 0.072.
  - Italy, 2000: input-output tables 0.444; national accounts 0.490; difference 0.046.
  - Netherlands, 2000: input-output tables 0.319; national accounts 0.315; difference -0.004.
  - Norway, 2001: input-output tables 0.342; national accounts 0.323; difference -0.018.
  - Poland, 2000: input-output tables 0.440; national accounts 0.400; difference -0.040.
  - Portugal, 1999: input-output tables 0.288; national accounts 0.361; difference 0.073.
  - Spain, 1995: input-output tables 0.267; national accounts 0.277; difference 0.011.
  - Sweden, 2000: input-output tables 0.435; national accounts 0.486; difference 0.051.
  - Australia, 1994/95: input-output tables 0.274; national accounts 0.361; difference 0.088.
  - Canada, 1997: input-output tables 0.324; national accounts 0.366; difference 0.042.
  - Czech Rep., 1995: input-output tables 0.488; national accounts 0.412; difference -0.076.
  - Greece, 1998: input-output tables 0.282; national accounts 0.349; difference 0.068.
  - Japan, 1995: input-output tables 0.280; national accounts 0.360; difference 0.081.
  - Korea, 1995: input-output tables 0.410; national accounts 0.378; difference -0.032.
  - UK, 1998: input-output tables 0.422; national accounts 0.499; difference 0.077.

### Appendix II — Two-sector model with tradables and nontradables (summary of analytical results)

- Model setup:
  - Consumption is aggregated as F(cT, cN) with consumption expenditure share on tradables denoted ε.
  - Foreign asset position b is exogenous and set equal to zero.
  - L denotes inelastic aggregate labor supply.
  - q is the relative price of capital goods.
- For the case of unitary elasticity of substitution in consumption (θ = 1), the model is characterized by a system of ten equations and ten unknowns (system (20)).
- Key solved relations and definitions:
  - From system (20) and equations 2, 3, 4, 5 and 9, expressions for A ratios, r, l, k, q, p, i are obtained as in (21), where γ1γ1γγ−−−=1G is set without loss of generality.
  - Equation (22) gives i in terms of parameters: () ,11 1 αα γγ δ α δ δα γ − ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ ++ − = NTNN AA rr LA i.
  - Equation (23) implies the analogous expression for the tradable sector: .1 1 α α γγ δ α δ δα γ − ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ ++ = − NTTT AA rr LA i.
- Labor and capital allocations:
  - Expressions for sectoral labor allocations and capital per sector are given in (24) and (25) (complex parameter-dependent expressions preserving ε, α, γ, δ, r, L, A terms).
- Output and investment:
  - Total output: ., ,1 α α γγ δ α − ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ + = ++ += − NTT NNNNTT AA r LA Y ipcpci Y (equation (26)).
  - Total investment: ., ,1 α α γγ δ α δ δα − ⎥ ⎦ ⎤ ⎢ ⎣ ⎡ ++ = += − NTT NNT AA rr LA I ipi I (equation (27)).
  - Investment rate: δ δα + = rY I (equation (28)).
  - Investment expenditure share on nontradables: () .1γ−= I ip NN (equation (29)).
- Productivity effects:
  - Total output is positively related to sectoral productivity parameters: () () .0 1 1 ,0 1 11 > − − = ∂ ∂ > − − − = ∂ ∂ NN TT A Y A Y A Y A Y α γα α γα (equation (30)).
- Comparisons in common prices (PPP adjustments):
  - Define pN PPP = AT/AN.
  - PPP adjusted investments expression (31) and PPP adjusted expenditure share on nontradables (32).
  - The derivative of the PPP nontradables share w.r.t. AT/AN is negative (equation (33)): higher relative productivity in the tradable sector leads to investment being less intensive in nontradables (same result as in (6)).
  - PPP adjusted output expressions (34) and derivatives (35) and (36) show ∂YPPP/∂AT > 0 and ∂YPPP/∂AN > 0 under the stated conditions.
  - Definitions of H1 and H2 in (37) are positive.
  - The PPP adjusted investment ratio is given in (38). The sign of ∂(I PPP / Y PPP)/∂(AT/AN) is given by (39): it is positive if εγ> and negative if εγ<; thus, investment rate increases (decreases) in relative productivity if expenditure share on tradables in investment is higher (lower) than in consumption.
- Allowing for CES aggregation in consumption:
  - If CES aggregator (18) with general θ is used, all expressions remain valid after the substitution in (40): .1 1 1 1 − − ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ + ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ − ⎟ ⎟ ⎠ ⎞ ⎜ ⎜ ⎝ ⎛ = θ θ μ μ ε N T A A.
  - Derivatives with respect to relative productivity change and need numerical evaluation when θ ≠ 1.

*Source: _wp0845 - Section 6 (IMF working paper PDF).*

### References

### _wp0845 - References

### Data sources and coverage
- OECD Input-Output database, ed. 1995: Input-output tables benchmark years for 1968-1990. Investment expenditure shares by industry. (OECD (1995a,b,c))
- OECD Input-Output database, ed. 2002: Input-output tables benchmark years for 1992-1998. Investment expenditure shares, tradability of sectoral output by industry. (OECD (2002a,b))
- Eurostat Input-Output database, ed. 2002: Input-output tables benchmark years for 1995-2005. Investment expenditure shares, tradability of sectoral output by industry. (Eurostat (2005))
- BEA Input-Output Table, 1997: Input-output tables 1997 benchmark (USA). Investment expenditure shares, tradability of sectoral output by industry. (Lawson et al. (2002))
- OECD annual detailed NA, GDP by expenditure, SNA 93, 1970-2004 (OECD (2006)) — used for investment expenditure shares and relative price of nontradables in investments across product categories: (i) products of agriculture, forestry, fishing and aquaculture [T], (ii) metal products and machinery [T], (iii) transport equipment [T], (iv) dwellings [N], (v) other buildings or structures [N] and (iv) other products [2/3 N, 1/3 T] /2/.
- Penn World Tables benchmarks: GDP by expenditure for 1970, 1975, 1980, 1985 and 1996 (Summers et al. (1995), Heston et al. (2002)) — used for investment expenditure shares, relative price of nontradables in investments, real income per capita; investment categories: (i) construction [N] and (ii) machinery and equipment [T]; before 1996 up to 20 subcategories of GFCF.
- Nehru-Dhareshwar dataset: GDP by expenditure 1987 (Nehru and Dhareshwar (1993)) — investment expenditure shares (i) construction[N] and (ii) machinery and equipment [T].
- UN annual detailed NA, GDP by expenditure, SNA 68, 1950-1997 (United Nations (2001a,b)) — investment expenditure shares across residential buildings [N], non-residential buildings [N], other construction and land development [N], other [T].

Notes from source text:
- 1. T and N in square brackets indicate classification into tradables [T] and nontradables [N].
- 2. Expenditures on "other products" largely overlap with expenditures on output of real estate/business services; defined as 2/3 nontradable and 1/3 tradable based on findings in Section 3.

### Investment expenditure shares by sector (selected tabulated results)
- Table 2 (OECD, 1968-1990 period averages and country-specific breakdowns):
  - Period average (Manufacturing, Construction, Retail/wholesale trade, Real est./bus. services, Other):
    - Manufacturing 0.338, Construction 0.561, Retail/wholesale trade 0.047, Real est./bus. services 0.028, Other 0.027 (period average row shows 0.338, 0.561, 0.047, 0.028, 0.027).
  - Country examples (country average column shown where present):
    - Australia country average: Manufacturing 0.294, Construction 0.584, Retail/wholesale trade 0.063, Real est./bus. services 0.024, Other 0.036.
    - Denmark country average: Manufacturing 0.346, Construction 0.573, Retail/wholesale trade 0.060, Real est./bus. services 0.020, Other 0.001.
    - France country average: Manufacturing 0.322, Construction 0.607, Retail/wholesale trade 0.028, Real est./bus. services --, Other 0.044.
    - Germany country average: Manufacturing 0.470, Construction 0.454, Retail/wholesale trade 0.037, Real est./bus. services 0.032, Other 0.006.
    - Italy country average: Manufacturing 0.393, Construction 0.516, Retail/wholesale trade 0.054, Real est./bus. services 0.017, Other 0.019.
    - Japan country average: Manufacturing 0.310, Construction 0.617, Retail/wholesale trade 0.064, Real est./bus. services 0.003, Other 0.007.
    - Netherlands country average: Manufacturing 0.352, Construction 0.488, Retail/wholesale trade 0.064, Real est./bus. services 0.066, Other 0.029.
    - UK country average: Manufacturing 0.434, Construction 0.447, Retail/wholesale trade 0.027, Real est./bus. services 0.059, Other 0.033.
    - USA country average: Manufacturing 0.367, Construction 0.532, Retail/wholesale trade 0.052, Real est./bus. services 0.031, Other 0.022.
  - Exact years of coverage for each country (selected): Australia - 1968, 1974, 1986, 1989; Denmark - 1972, 1977, 1980, 1985, 1990; France - 1972, 1977; Germany - 1978, 1986, 1988, 1990; Italy - 1985; Japan - 1970, 1975, 1980, 1985, 1990; Netherlands - 1972, 1977, 1981, 1986; UK - 1968, 1979, 1984, 1990; United States - 1972, 1977, 1982, 1985, 1990.
  - Data covers only private sector investment. Expenditures on real estate/business services are included in ‘other’ sectors. Data for Canada and France after 1980 excluded because GFCF in input-output tables covered only a fraction of aggregate investment.

- Table 3 (USA, 1997, BEA input-output benchmark table):
  - Manufacturing 0.354
  - Construction 0.426
  - Retail and wholesale trade 0.073
  - Real estate and business services 0.121
    - of which: Real estate services 0.028
    - Computer system design and related services 0.056
    - Software 0.023
    - Other business services 0.014
  - Other 0.026
  - Note: To make results compatible with SNA93 definitions, investment includes private and government investment, except military expenditures.

- Table 4 (Eurostat and national IO tables, investment expenditure shares, 1998-2001 for selected countries):
  - Selected country-year entries:
    - UK 1998: Manufacturing 0.422; Construction 0.410; Retail/wholesale trade 0.048; Real est./bus. services 0.090; Other 0.030.
    - Poland 2000: Manufacturing 0.440; Construction 0.370; Retail/wholesale trade 0.092; Real est./bus. services 0.084; Other 0.015.
    - Norway 2001: Manufacturing 0.342; Construction 0.329; Retail/wholesale trade 0.051; Real est./bus. services 0.191; Other 0.088.
    - Italy 2000: Manufacturing 0.444; Construction 0.401; Retail/wholesale trade 0.063; Real est./bus. services 0.068; Other 0.024.
    - France 2000: Manufacturing 0.339; Construction 0.436; Retail/wholesale trade 0.036; Real est./bus. services 0.174; Other 0.016.
    - Netherlands 2000: Manufacturing 0.319; Construction 0.428; Retail/wholesale trade 0.074; Real est./bus. services 0.117; Other 0.063.
    - Ireland 1998: Manufacturing 0.279; Construction 0.568; Retail/wholesale trade 0.082; Real est./bus. services 0.065; Other 0.007.
    - Average (across listed countries): Manufacturing 0.367; Construction 0.438; Retail/wholesale trade 0.066; Real est./bus. services 0.105; Other 0.025.

### Tradability of sectoral output (Table 5)
- Definition: Tradability = (imports + exports)/gross output. 'N' indicates sector's tradability lower than retail/wholesale trade sector in same country; 'T' indicates tradable.
- Selected country-sector tradability values (sector columns: Retail and wholesale trade, Construction, Real estate and business services, Manufacturing):
  - USA, 1997: Retail/wholesale trade 0.04; Construction 0.00; Real estate and business services N 0.02; Manufacturing 0.29 T.
  - France, 2000: Retail/wholesale trade 0.10; Construction 0.00; Real estate and business services N 0.07; Manufacturing 0.54 T.
  - Germany, 2000: Retail/wholesale trade 0.09; Construction 0.02; Real estate and business services N 0.06; Manufacturing 0.62 T.
  - Austria, 2000: Retail/wholesale trade 0.13; Construction 0.04; Real estate and business services N 0.13; Manufacturing 0.77 T.
  - Belgium, 2000: Retail/wholesale trade 0.24; Construction 0.03; Real estate and business services N 0.22; Manufacturing 0.92 T.
  - Ireland, 1998: Retail/wholesale trade 0.51; Construction 0.00; Real estate and business services N 0.59 T; Manufacturing 0.90 T.
  - China, 1997: Retail/wholesale trade 0.11; Construction 0.00; Real estate and business services N 0.02; Manufacturing 0.21 T.
  - Japan, 1995: Retail/wholesale trade 0.03; Construction 0.00; Real estate and business services N 0.02; Manufacturing 0.19 T.
- Sample mean/median across listed countries by sector:
  - Retail/wholesale trade mean/median 0.12/0.09
  - Construction mean/median 0.02/0.00
  - Real estate and business services mean/median 0.12/0.09
  - Manufacturing mean/median 0.58/0.60

### Aggregate investment on nontradable goods (Table 6)
- Summary of aggregate investment expenditures on nontradable goods, 1970-2000 (expenditure share):
  - Pre-1973 0.64
  - Mid/late-1970s 0.63
  - Early-1980s 0.63
  - Mid-1980s 0.59
  - 1990 0.61
  - Mid-1990s 0.61
  - 2000 0.59
- Methodology note: Until 1990 expenditure share obtained from period averages in Table 2, assuming expenditures on construction sector, distribution sector and 2/3 of real estate business sector are expenditures on nontradable goods. Results for 2000 and mid-1990s obtained by applying same definitions to Table 4 and relevant IO data (including U.S. Table 3 for mid-1990s).

### Summary statistics for investment expenditures on nontradable goods, OECD data (Table 7)
- Table columns: Country; Years of coverage; Mean; Standard deviation; Max; Min; Time trend, per decade/1/; Newey-West standard error.
- Selected country entries (preserving all numeric values exactly as listed):
  - Australia 1970-2004: Mean 0.61; Standard deviation 0.032; Max 0.66; Min 0.52; Time trend, per decade/1/ 0.025*; Newey-West standard error 0.006.
  - Austria 1976-2004: Mean 0.59; Standard deviation 0.019; Max 0.63; Min 0.57; Time trend 0.005; Newey-West standard error 0.005.
  - Canada 1970-2004: Mean 0.66; Standard deviation 0.031; Max 0.70; Min 0.59; Time trend -0.024*; Newey-West standard error 0.006.
  - Denmark 1970-2004: Mean 0.59; Standard deviation 0.056; Max 0.70; Min 0.52; Time trend -0.046*; Newey-West standard error 0.010.
  - Finland 1970-2004: Mean 0.62; Standard deviation 0.026; Max 0.68; Min 0.58; Time trend 0.008; Newey-West standard error 0.007.
  - Germany 1970-2004: Mean 0.61; Standard deviation 0.029; Max 0.67; Min 0.56; Time trend -0.008; Newey-West standard error 0.006.
  - Italy 1970-2004: Mean 0.53; Standard deviation 0.029; Max 0.60; Min 0.48; Time trend -0.018*; Newey-West standard error 0.006.
  - Japan 1980-2003: Mean 0.61; Standard deviation 0.020; Max 0.66; Min 0.59.
  - Netherlands 1970-2004: Mean 0.64; Standard deviation 0.031; Max 0.69; Min 0.58; Time trend -0.011; Newey-West standard error 0.008.
  - Norway 1970-2004: Mean 0.67; Standard deviation 0.046; Max 0.75; Min 0.55; Time trend 0.027*; Newey-West standard error 0.011.
  - Spain 1980-2004: Mean 0.64; Standard deviation 0.030; Max 0.70; Min 0.60.
  - Sweden 1980-2004: Mean 0.53; Standard deviation 0.049; Max 0.61; Min 0.44.
  - United Kingdom 1970-2004: Mean 0.53; Standard deviation 0.029; Max 0.60; Min 0.47; Time trend -0.001; Newey-West standard error 0.009.
  - United States 1970-2004: Mean 0.60; Standard deviation 0.023; Max 0.65; Min 0.56; Time trend -0.010; Newey-West standard error 0.008.
- Other country entries included in Table 7 with exact numeric values: Czech Republic 1990-2003 Mean 0.54 Std dev 0.054 Max 0.62 Min 0.46; Greece 1995-2004 Mean 0.62 Std dev 0.027 Max 0.67 Min 0.59; Hungary 2000-2004 Mean 0.54 Std dev 0.026 Max 0.57 Min 0.50; Iceland 1990-2004 Mean 0.67 Std dev 0.049 Max 0.76 Min 0.60; Ireland 1990-2004 Mean 0.65 Std dev 0.062 Max 0.77 Min 0.56; Korea 1970-2004 Mean 0.59 Std dev 0.053 Max 0.71 Min 0.51 Time trend 0.021 Newey-West standard error 0.013; Luxembourg 1985-2003 Mean 0.58 Std dev 0.051 Max 0.70 Min 0.51; New Zealand 1971-2004 Mean 0.55 Std dev 0.033 Max 0.59 Min 0.48 Time trend 0.000 Newey-West standard error 0.009; Poland 1995-2003 Mean 0.58 Std dev 0.021 Max 0.62 Min 0.56; Portugal 1988-2004 Mean 0.59 Std dev 0.029 Max 0.64 Min 0.54; Slovak Republic 1993-2004 Mean 0.42 Std dev 0.047 Max 0.48 Min 0.35; Switzerland 1990-2003 Mean 0.50 Std dev 0.025 Max 0.55 Min 0.47.
- Average across OECD sample: Mean 0.59; Standard deviation 0.035; Max 0.65; Min 0.53.

* indicates statistical significance for the reported time trend where annotated in the table.

_Source: _wp0845 - References_

### 1. Calculated for countries with at least 30 annual observations. The time trend reports the estimate

### 1. Calculated for countries with at least 30 annual observations. The time trend reports the estimate

### Methodology
- Time trend reports the estimate of 10*β from the regression: γ_t = α + β t + ε_t, where t denotes years.
- * indicates that the time trend is significantly different from zero at the 5% confidence level.
- The slope of expenditure shares is multiplied by 10 and should therefore be interpreted as a change in expenditure share over a decade.
- The N-W (Newey-West) standard error is also multiplied by 10.
- Data source: UN (2001a, b).

### Table 8 — Summary statistics (investment expenditures on nontradable goods, UN data)
- Table fields per country: Years of coverage; Mean; Standard deviation; Max; Min; Time trend, per decade/1/; Newey-West standard error.
- Many country entries do not report a time trend and Newey-West standard error in the table excerpt; where reported, both values are shown.

### Selected notable country statistics (preserving exact reported values)
- Australia (1959-96): Mean 0.54; Standard deviation 0.021; Max 0.60; Min 0.49; Time trend, per decade/1/ -0.006; Newey-West standard error 0.004
- Austria (1954-96): Mean 0.54; Standard deviation 0.024; Max 0.58; Min 0.47; Time trend, per decade/1/ 0.015*; Newey-West standard error 0.005
- Belgium (1960-97): Mean 0.59; Standard deviation 0.050; Max 0.68; Min 0.49; Time trend, per decade/1/ -0.026*; Newey-West standard error 0.010
- Canada (1950-97): Mean 0.68; Standard deviation 0.025; Max 0.73; Min 0.63; Time trend, per decade/1/ -0.009; Newey-West standard error 0.005
- Hong Kong (1961-97): Mean 0.40; Standard deviation 0.041; Max 0.50; Min 0.30; Time trend, per decade/1/ -0.000; Newey-West standard error 0.006
- Colombia (1960-95): Mean 0.58; Standard deviation 0.041; Max 0.66; Min 0.49; Time trend, per decade/1/ -0.022*; Newey-West standard error 0.005
- Cyprus (1960-96): Mean 0.65; Standard deviation 0.059; Max 0.74; Min 0.53; Time trend, per decade/1/ 0.041*; Newey-West standard error 0.006
- Denmark (1966-95): Mean 0.60; Standard deviation 0.066; Max 0.70; Min 0.49; Time trend, per decade/1/ -0.070*; Newey-West standard error 0.007
- Guatemala (1950-96): Mean 0.41; Standard deviation 0.115; Max 0.67; Min 0.26; Time trend, per decade/1/ -0.065*; Newey-West standard error 0.007
- India (1950-96): Mean 0.56; Standard deviation 0.078; Max 0.75; Min 0.42; Time trend, per decade/1/ -0.053*; Newey-West standard error 0.003
- Germany, Federal Rep. of (1960-94): Mean 0.61; Standard deviation 0.035; Max 0.66; Min 0.53; Time trend, per decade/1/ -0.025*; Newey-West standard error 0.007
- Greece (1960-95): Mean 0.63; Standard deviation 0.050; Max 0.72; Min 0.53; Time trend, per decade/1/ -0.043*; Newey-West standard error 0.003
- Israel (1950-97): Mean 0.61; Standard deviation 0.089; Max 0.83; Min 0.43; Time trend, per decade/1/ -0.052*; Newey-West standard error 0.006
- Italy (1960-97): Mean 0.56; Standard deviation 0.047; Max 0.65; Min 0.48; Time trend, per decade/1/ -0.032*; Newey-West standard error 0.009
- Lesotho (1964-96): Mean 0.66; Standard deviation 0.125; Max 0.89; Min 0.45; Time trend, per decade/1/ 0.093*; Newey-West standard error 0.021
- Norway (1960-96): Mean 0.59; Standard deviation 0.056; Max 0.66; Min 0.47; Time trend, per decade/1/ 0.046*; Newey-West standard error 0.009
- Paraguay (1962-94): Mean 0.54; Standard deviation 0.086; Max 0.78; Min 0.43; Time trend, per decade/1/ 0.040*; Newey-West standard error 0.014
- Philippines (1950-97): Mean 0.54; Standard deviation 0.083; Max 0.71; Min 0.37; Time trend, per decade/1/ -0.032*; Newey-West standard error 0.006
- Puerto Rico (1950-96): Mean 0.63; Standard deviation 0.072; Max 0.76; Min 0.50; Time trend, per decade/1/ -0.037*; Newey-West standard error 0.005
- South Africa (1963-97): Mean 0.53; Standard deviation 0.057; Max 0.60; Min 0.39; Time trend, per decade/1/ -0.036*; Newey-West standard error 0.003
- Sri Lanka (1963-97): Mean 0.61; Standard deviation 0.086; Max 0.74; Min 0.41; Time trend, per decade/1/ -0.059*; Newey-West standard error 0.006
- Switzerland (1950-96): Mean 0.61; Standard deviation 0.046; Max 0.68; Min 0.53; Time trend, per decade/1/ -0.022*; Newey-West standard error 0.006
- United Kingdom (1963-96): Mean 0.54; Standard deviation 0.021; Max 0.59; Min 0.50; Time trend, per decade/1/ -0.003; Newey-West standard error 0.005
- United States (1960-97): Mean 0.59; Standard deviation 0.040; Max 0.66; Min 0.52; Time trend, per decade/1/ -0.034*; Newey-West standard error 0.003
- Uruguay (1966-89): Mean 0.69; Standard deviation 0.076; Max 0.82; Min 0.58; Time trend, per decade/1/ 0.061; Newey-West standard error 0.030

### Aggregate summary from the table
- Average   0.58 0.061 0.69     0.48

*Data source: UN (2001a, b).*

### 1. See notes to Table 7.

### _wp0845 - 1. See notes to Table 7.

### Correlation of expenditure shares on nontradables (Table 9)
- Pairwise correlation matrix of expenditure shares on nontradables (UN data) across decades:
  - # of countries included: 13 for 1950-59, 44 for 1960-69, 91 for 1970-79, 91 for 1980-89, 80 for 1990-97.
  - Correlations reported by decade (row/column ordering follows sample periods shown):
    - 1950-59 vs 1950-59: 1
    - 1960-69 vs 1950-59: 0.6491
    - 1970-79 vs 1950-59: 0.4520.8631 (as shown in source table)
    - 1980-89 vs 1950-59: 0.5270.5550.6431 (as shown in source table)
    - 1990-97 vs 1950-59: 0.3070.6300.4920.7351 (as shown in source table)
- Data sources: UN (2001a, b).

### Pooled time trends in nontradable expenditure shares (Table 10)
- Estimation approaches and samples:
  - Panel 1: OECD data (Pooled OLS)
    - Sample: OECD countries
    - # of observations: 448
    - Time trend, per decade: -0.003
    - Standard error: 0.003
  - Panel 1: OECD data (Panel with country dummies)
    - Sample: OECD countries
    - # of observations: 448
    - Time trend, per decade: -0.003
    - Standard error: 0.002
  - Panel 2: UN data (Pooled OLS)
    - Sample: all countries
    - # of observations: 1335
    - Time trend, per decade: -0.014* 
    - Standard error: 0.002
    - Sample: OECD countries
      - # of observations: 610
      - Time trend, per decade: -0.014*
      - Standard error: 0.002
    - Sample: Non-OECD countries
      - # of observations: 725
      - Time trend, per decade: -0.016*
      - Standard error: 0.003
  - Panel 2: UN data (Panel with country dummies)
    - Sample: all countries
      - # of observations: 1335
      - Time trend, per decade: -0.017*
      - Standard error: 0.002
    - Sample: OECD countries
      - # of observations: 610
      - Time trend, per decade: -0.012*
      - Standard error: 0.001
    - Sample: Non-OECD countries
      - # of observations: 725
      - Time trend, per decade: -0.020*
      - Standard error: 0.003
- Notes:
  - In pooled OLS, time trend reports estimate of 10*β from γt = α+βt+εt.
  - In panel with country dummies, time trend reports estimate of 10*β from γt = α+βt+di+εt.
  - * indicates significance at the 5% confidence level.
  - Time trend slopes and N-W standard errors are multiplied by 10 and interpreted as change in expenditure share over a decade.
- Data sources: OECD (2006), UN (2001a, b).

### Cross-section comparison: investment expenditures on nontradable goods (UN data, 1950-1997) (Table 11)
- Annual cross-section summary statistics and regression results (selected lines and overall averages):
  - 1950: # of countries included 9; Mean 0.68; Correlation with real income per capita -0.31; OLS coefficient -0.030; Robust standard error 0.014
  - 1959: # of countries included 13; Mean 0.61; Correlation with real income per capita 0.04; OLS coefficient -0.002; Robust standard error 0.021
  - 1960: # of countries included 30; Mean 0.60; Correlation with real income per capita 0.07; OLS coefficient 0.005; Robust standard error 0.016
  - 1969: # of countries included 44; Mean 0.58; Correlation with real income per capita 0.27; OLS coefficient 0.028; Robust standard error 0.014
  - 1970: # of countries included 71; Mean 0.56; Correlation with real income per capita 0.28; OLS coefficient 0.022*; Robust standard error 0.010
  - 1971: # of countries included 76; Mean 0.56; Correlation with real income per capita 0.32*; OLS coefficient 0.024*; Robust standard error 0.009
  - 1978: # of countries included 81; Mean 0.56; Correlation with real income per capita 0.30*; OLS coefficient 0.025*; Robust standard error 0.012
  - 1987: # of countries included 79; Mean 0.55; Correlation with real income per capita 0.18*; OLS coefficient 0.029*; Robust standard error 0.013
  - 1990: # of countries included 74; Mean 0.54; Correlation with real income per capita 0.26*; OLS coefficient 0.034*; Robust standard error 0.016
  - 1995: # of countries included 59; Mean 0.57; Correlation with real income per capita -0.15; OLS coefficient -0.027; Robust standard error 0.026
  - 1996: # of countries included 48; Mean 0.58; Correlation with real income per capita -0.23; OLS coefficient -0.040; Robust standard error 0.025
  - 1997: # of countries included 21; Mean 0.52; Correlation with real income per capita 0.08; OLS coefficient 0.019; Robust standard error 0.037
  - Average across years: Mean 0.58; Average correlation with real income per capita 0.10
- Notes:
  - Correlation coefficient marked * indicates significance at 5% confidence level.
  - For each year OLS coefficient and robust standard error from regression of nontradable investment share on log of real income per capita; * indicates coefficient significant at 5% confidence level.
- Data sources: UN (2001a, b), real GDP data from Heston et al. (2002).

### Cross-section comparison: PWT benchmark data (Table 12)
- Investment expenditure on nontradables by data set:
  - PWT 1996 benchmark/1/: # of countries included 115; Mean 0.51; Correlation with real income per capita 0.12
    - -only A,B: 33 countries; Mean 0.56; Correlation 0.03
    - -only A: 18 countries; Mean 0.56; Correlation 0.10
  - Nehru-Dhareshwar dataset, 1987: 42 countries; Mean 0.56; Correlation 0.13
  - PWT 1985 benchmark: 65 countries; Mean 0.56; Correlation 0.04
  - PWT 1980 benchmark: 60 countries; Mean 0.58; Correlation 0.31
  - PWT 1975 benchmark: 34 countries; Mean 0.57; Correlation 0.53
  - PWT 1970 benchmark: 16 countries; Mean 0.56; Correlation 0.13
- Note: A, B, C and D refer to data quality categories in Penn World Table 6.1 benchmark.
- Sources: Heston et al. (2002), Summers et al. (1995).

### Regional patterns in investment shares on nontradables (Table 13)
- PWT 1996 benchmark, average expenditure share by region:
  - Western Europe and North America/1/: # of countries included 25; Average expenditure share 0.56
  - Africa: # of countries included 22; Average expenditure share 0.23
    - Africa, PWT 1985: 22 countries; 0.54
    - Africa, PWT 1980: 15 countries; 0.57
  - Eastern and Central Europe/2/: # of countries included 26; Average expenditure share 0.59
  - Asia: # of countries included 12; Average expenditure share 0.59
  - Caribbean: # of countries included 12; Average expenditure share 0.51
  - Latin America: # of countries included 10; Average expenditure share 0.57
  - Middle East: # of countries included 8; Average expenditure share 0.57
- Notes:
  - Western Europe and North America includes Japan, Australia and New Zealand.
  - Eastern and Central Europe includes all former republics of the Soviet Union.
- Source: Heston et al. (2002).

### Comparison of construction investment shares: Burstein et al. (2004) vs OECD national accounts (Table 14)
- Country-level examples (Real GDP/capita; Burstein et al. (2004) expenditure share; OECD NA data expenditure share):
  - Korea 1993: Real GDP/capita 11940; Burstein 0.540; OECD NA 0.639
  - Mexico 1990: Real GDP/capita 7429; Burstein 0.485; OECD NA 0.502
  - Brazil 1999: Real GDP/capita 6909; Burstein 0.674; OECD NA 0.669
  - Argentina 1997: Real GDP/capita 11349; Burstein 0.542; OECD NA 0.635
  - Australia 1995: Real GDP/capita 22164; Burstein 0.500; OECD NA 0.566
  - Canada 1990: Real GDP/capita 22427; Burstein 0.526; OECD NA 0.639
  - Chile 1996: Real GDP/capita 8972; Burstein 0.596; OECD NA 0.510
  - Denmark 1998: Real GDP/capita 25495; Burstein 0.457; OECD NA 0.469
  - Finland 1995: Real GDP/capita 18852; Burstein 0.458; OECD NA 0.524
  - France 1995: Real GDP/capita 20142; Burstein 0.485; OECD NA 0.498
  - Germany 1995: Real GDP/capita 21049; Burstein 0.494; OECD NA 0.640
  - Greece 1996: Real GDP/capita 12751; Burstein 0.647; OECD NA 0.645
  - Italy 1992: Real GDP/capita 19810; Burstein 0.498; OECD NA 0.498
  - Japan 1995: Real GDP/capita 23361; Burstein 0.573; OECD NA 0.552
  - Netherlands 1996: Real GDP/capita 21431; Burstein 0.432; OECD NA 0.538
  - Norway 1997: Real GDP/capita 26178; Burstein 0.346; OECD NA 0.409/2/
  - Spain 1995: Real GDP/capita 16296; Burstein 0.564; OECD NA 0.572
  - UK 1998: Real GDP/capita 21693; Burstein 0.410; OECD NA 0.427
  - US 1997: Real GDP/capita 30286; Burstein 0.423; OECD NA 0.485
- Correlation of expenditure shares with real GDP/capita:
  - Correlation with real GDP/capita (Burstein et al. (2004)): -0.69
  - Correlation with real GDP/capita (OECD NA data): -0.49
- Notes:
  - For non-OECD countries, investment shares obtained from PWT 1996 benchmark data.
  - Norway OECD NA figure excludes oil rigs and oil exploration related expenditures (note /2/).
- Sources: Burstein et al. (2004), OECD (2006), Heston et al. (2002).

### Relative prices of nontradables in investment (Figures 5 and 6)
- Cross-country scatter of relative price of nontradables vs Real GDP per capita (1996 PWT benchmark; Figure 5)
  - Caribbean islands excluded from figure because each exhibited higher relative price than any other country; average relative price for Caribbean islands is 3.74.
  - Source: Heston et al. (2002).
- Time series for selected countries (Figure 6)
  - Relative price of nontradables in investment plotted for FR A G ER I TA J P N U K USA over years 1970–2004 (OECD (2006) source).

### Regional and temporal patterns (Figures 1–4)
- Figure 1 (OECD data): Time series of investment expenditure share on nontradables for selected countries (FRAGERITAJPNUKUSA) over 1970–2004; source OECD (2006).
- Figure 2: Cross-section averages of investment share on nontradables excluding residential construction (OECD and UN data), shown over 1960–2002; sources OECD (2006) and UN (2001a, b).
- Figure 3: Scatterplots of investment expenditure share on nontradables versus real GDP per capita for years 1960, 1970, 1980, 1990; reported correlations:
  - 1960: Corr. = 0.076
  - 1970: Corr. = 0.283
  - 1980: Corr. = 0.076
  - 1990: Corr. = 0.266
  - Sources: UN (2001a, b) and Heston et al. (2002).
- Figure 4: Regional time series of investment expenditure share on nontradables for Africa, Europe, Latin America, South East Asia, and All over 1960–1996; source UN (2001a, b).

### Model sensitivity and variation in investment rates (Figure 7)
- Variation in investment rates under different model specifications plotted against real income per capita (relative to average for countries with at least 75% of U.S. income):
  - Scenarios/lines shown:
    - γ = 1; μ = 0
    - γ = 0.4; θ = 0.44
    - γ = 0.4; μ = 0
    - linear trend in data
  - X-axis: real income per capita (relative to the average for countries with at least 75% of the U.S. income).
  - Y-axis: investment rates in international prices.
  - Data source: PWT 1996 benchmark (Heston et al. (2002)).

*Source: Content unit _wp0845 - 1. See notes to Table 7.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2008/_wp0845.pdf_