WPS 256. POLICY RESEARCH WORKING PAPER 2552 Inventories in Developing High inventory levels in developing countries increase Countries the cost of doing business and limit productivity and Levels and Determinants-a Red Flag competitiveness. Improvements in for Competitiveness and Growth infrastructure (roads, ports, and telecommunications) and J. Lutis Gpascb in market development can Joseph Kogani help to significantly reduce inventory levels (and thus the cost of doing business), especially when accompanied by effective regulation and the development and deregulation of associated markets. The World Bank Development Economics Office of the Senior Vice President and Chief Economist and Latin America and the Caribbean Region Finance, Private Sector, and Infrastructure Sector Management Unit February 2001 H POLICY RESEARCH WORKING PAPER 2552 Summary findings Raw materials inventories in the manufacturing sector in materials inventories by 27-47 percent. Poorly the 1970s, 1980s, and 1990s were two to five times as functioning markets, as measured by the ratio of high in developing countries as in the United States, transfers and subsidies to GDP, are also an important despite the fact that in most developing countries real factor, with a one-standard-deviation improvement interest rates are at least twice as high. Given the high leading to a 19-30 percent reduction in raw materials cost of capital in most developing countries, these high inventories. inventory levels have an enormous impact on the cost of Guasch and Kogan show that these reductions in raw doing business and on productivity and competitiveness. materials inventories are not offset by a reduction in Poor infrastructure and ineffective regulation as well as finished goods inventories upstream. deficiencies in market development-rather than the The policy implications are clear and strong. traditional factors used in inventory models, such as Improvements in infrastructure (roads, ports, and interest rates and uncertainty-are the main telecommunications) can help to significantly reduce determinants of these differences. inventory levels (and thus the cost of doing business), Cross-country estimates show that a one-standard- especially when accompanied by effective regulation and deviation improvement in infrastructure reduces raw the development and deregulation of associated markets. This paper-a joint product of the Office of the Senior Vice President and Chief Economist, Development Economics, and the Finance, Private Sector, and Infrastructure Sector Management Unit, Latin America and the Caribbean Region-is part of a larger effort in the Bank to assess and improve the competitiveness and productivity of developing countries. Copies of the paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Joy Troncoso, room 15-062, telephone 202-473-7826, fax 202-522-2106, email address jtroncoso@worldbank.org. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at jguasch@worldbank.org or jkogan@fas.harvard.edu. February 2001. (25 pages) The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Produced by the Policy Research Dissemination Center Inventories in Developing Countries: Levels and Determinants--a Red Flag for Competitiveness and Growth J. Luis Guasch Latin America and Caribbean (LAC) and Development Economics and Chief Economist (DEC), World Bank and University of California, San Diego Jguasch@worldbank.org and Joseph Kogan Harvard University Jkogan@fas.harvard.edu We thank Abhijit Banerjee, Tarun Khanna, Michael Kremer, Andrei Shleifer, Joseph Stiglitz and Harvard workshop participants for helpful comments on this paper. We also thank Lis Suarez for assistance with data entry. Partial funding from PPIAF and Harvard University is gratefully acknowledged. The views in this paper reflect those of the authors and do not necessarily reflect the views of the institutions with which they are affiliated. Table of Contents 1. Introduction ...................................... 4 2. Theoretical Overview ...................................... 6 3. Data Description ...................................... 11 4. Analysis of Determinants of Raw Material Inventory ...................................... 14 5. Input-Output Analysis ...................................... 21 6. Conclusion ........................................................................................................ 23 1. Introduction Although it is well known from anecdotal evidence that inventories are higher in developing countries, there are almost no systematic studies that attempt to explain this phenomenon or even to quantify the difference. This study uses newly-assembled data for 52 countries in the early 1970s and 1980s to draw out some stylized facts about the pattern of inventory holdings, and more recent data in Latin American countries for the 1990s shows that the problem persist. The motivation for this paper is the magnitude of the holdings and the potential cost to the economy. U.S. businesses typically hold inventories equal to about 15% of GDP while inventory levels in many developing countries are often twice as large and for raw materials three times as large (Table 1). If the private sector interest rate for financing inventory holdings is 15%-20%, a conservative estimate in most developing countries, then the cost to the economy of the additional inventory holdings is greater than 2% of GDP. Suppose that firms in developing countries keep high levels of inventories in response to poor infrastructure, which we find in this study to be a key determinant. Then, as an example, consider that the total transport infrastructure stock in Bangladesh is about 2% of GDP' (World Bank 1994) while this figure is about 12% in the United States.2 One year's worth of savings in inventory holding costs would be enough to double Bangladesh's infrastructure stock; the infrastructure improvement could pay for itself. At the firm level, the impact of these high levels of inventories is also enormous. Given the high cost of capital in many developing countries, cutting inventory levels in half 1 Rough calculation based on graphs of infrastructure stock per capita and composition of infrastructure. See World Bank [1994], figures 1 and 2. 2Nonmilitary nonresidential net public stock in the United States in 1991 was $2.2 trillion with $700 billion of this amount representing stocks of highways and streets. See Munnell [1992], p. 190 for U.S. infrastructure data. could reduce unit costs by over 20%, with a significant impact on competitiveness, aggregate demand, and employment. Table 1: Latin America Ratios to U.S. Inventories (all industries) S;0~ ~ ~ ~~~a Maeral lamatory LeveR RatIo: Rai to X.S LeveR000 by;0 0 07R ;v lvX '- -- | Chile Venezuela Peru Bolivia Colombia Ecuador Mexico Brazil Mean 2.17 2.82 4.19 4.20 2.22 5.06 1.58 2.98 Minimum 0.00 0.30 0.10 0.11 0.52 0.86 0.42 0.8 Ist Quartile 0.36 1.87 1.25 1.39 1.45 2.55 1.06 1.6 Median 1.28 2.61 2.30 2.90 1.80 3.80 1.36 2.00 3rd Quartile 2.66 3.12 3.90 4.49 2.52 5.64 2.06 3.1 Maximum 68.92 7.21 31.1 34.97 13.59 20.61 3.26 7.1 _________ _ (: :rage of D s0 Falab:le data _t_ to_ _ , == _______ Chile Venezuela Peru Bolivia Colombia Ecuador Mexico Brazil Mean 1.76 1.63 1.65 2.74 1.38 2.57 1.46 1.98 Minimum 0.01 0.10 0.39 0.11 0.19 0.67 0.35 0.75 1st Quartile 0.17 0.87 1.17 1.13 1.05 1.67 0.82 1.1 Median 0.72 1.60 1.54 2.02 1.28 1.98 1.36 1.60 3rd Quartile 1.38 2.1.4 2.11 3.18 1.63 2.86 12.14 2.00 Maximum 31.61 5.29 3.87 21.31 5.31 7.94 4.91 5.2 Source: Guasch and Kogan (2000) These calculations are merely a lower bound on the cost of the additional inventory. First, there are certain transactions that would have been worthwhile were A not for the high level of inventory holdings necessary to complete them effectively. It is difficult to estimate the size of these lost transactions. Second, firms in developing countries will take costly steps to mitigate the institutional or structural factors creating a need for high inventories. Suppose that for a particular firm, 30 days of inventory are sufficient when transportation networks are well developed but 90 days of inventory are required when transportation networks are poor. The firm might choose to reduce these 90 days to 60 days by requiring suppliers to locate nearby. Additional costs due to poor infrastructure as measured by increased inventory levels would be 5 30 days while the actual costs are higher.3 Third, high inventories can obscure efficiency problems. Current thinking in the manufacturing and operations research field suggests that low inventories make it easier to trace problems in the production process.4 The objective of this paper is to systematically report the high levels of inventories in developing countries and to impute their determinants, pointing to policy interventions to considerably reduce those levels. Section 2 of this paper provides a brief theoretical overview of why firms hold inventory and why developing countries might hold more. Section 3 describes the data we have collected. Section 4 contains the estimations which show that inventory levels are significantly higher in developing countries due to poor infrastructure and market interference. Section 5 checks the data for shifting of inventories to upstream industries. Section 6 concludes. 2. Theoretical Overview The economics literature typically cites three theoretical reasons for why businesses hold inventory: production smoothing, stockout avoidance, and reduction of transaction costs. Blinder [1991] gives examples of other reasons such as holding inventories for display purposes or to speculate on or hedge against price movements, but the above three explanations are the most prevalent. The mathematical modeling of optimal inventory policies is a field in itself, with 3 Gulyani [2000] describes how Maruti, an Indian automaker, tries to decrease inventory costs by encouraging its suppliers to locate nearby through govermnent-sponsored incentive packages and the building of supplier parks. Fisman and Khanna [1998] describes co-location by business group affiliates to overcome infrastructure shortages. 4Nahmias [1997], p. 373, states in a discussion of just-in-time inventory management, "A popular analogy is to compare a production process with a river and the level of inventory with the wate level in the river. When the water level is high, the water will cover the rocks. Likewise, when inventory levels are high, problems are masked. However, when the water levels (inventory) is low, the rocks (problems) are evident. Because items are moved through the system in small batches, 100 percent inspection is feasible. Seen in this light, just-in-time can be easily incorporated into an overall quality control strategy." 6 much work done by economists, mathematicians, and operations researchers.S Here, we merely describe the three reasons intuitively. In the production smoothing model, firms have a rising marginal cost curve. Firms seeking to minimize production costs in the face of sales that vary predictably over time will produce a constant amount every month, accumulating inventories when sales are below production and depleting inventories when sales exceed production. Firms select their inventory levels by weighing storage and financing costs against potential savings from production optimization.6 The stockout motive presumes that demand varies unpredictably over time and any demand that cannot be satisfied immediately out of inventory will be lost rather than carried over into the next period. Firms hold inventory to meet this unanticipated demand. While the production smoothing motive only explains why manufacturers would keep finished goods inventories, the stockout motive explains the existence of retail inventories and raw materials inventories as well. The stockout motive applies also if the uncertainty occurs not in demand but in the timing of deliveries. Firms concerned about stockout optimize inventory levels by trading off holding costs against the likelihood of stockout. 5 Fafchamps et al. [19971 provides some simple mathematical models. For some work by economists on this subject, see Arrow, Karlin, and Scarf [1958], Scarf, Gilford, and Shelly [1963], and Scarf [1960]. For articles by mathematicians, see issues of Siam Journal of Applied Mathematics. Nahmias [19971 is a commonly used textbook for studying production operations management, covering a number of basic models and providing numerous academic references. 6 The production smoothing motive does not appear to hold empirically. Blinder [1991] cites three basic facts about U.S. inventories which seem to discredit the production smoothing explanation (1) Production is more variable than sales in most industries. (2) Sales and inventory investment normally are not negatively correlated. (3) The most volatile components of inventory investment are retail inventories and manufacturers' inventories of raw materials and supplies while production smoothing only applies to finished goods inventories. Economists have attempted to reconcile these facts with production smoothing by introducing cost shocks but these explanations have not been empirically successful. Fukuda and Teruyama [1988], however, show that the stylized fact that production is more variable than sales is representative of developed economies but not of developing economies. 7 The transaction cost motive assumes that there are certain fixed costs to placing an order or that there are economies of scale in ordering in large batches. When faced with uncertain demand as in the stockout model, firms follow an (S,s) strategy. As soon as the inventory falls below s, the firm places an order of a lot size equal to S-s so that the inventory level for each firm fluctuates between s and S. In determining the optimal lot size, firms weigh inventory holding costs against savings from large orders. According to Mosser [1991], retail inventories are usually managed by an (S,s) rule, as evidenced by its presence in textbooks on purchasing, retailing, and merchandising as well as in trade journals and business reviews which describe implementations of the (S,s) rule using computers.7 Poor infrastructure would affect raw materials inventories either through the stockout or transaction cost motive. According to the stockout model, poor infrastructure could increase the time it takes for a shipment to arrive. When a firm finds itself running low on raw materials due to a sudden increase in demand for its finished products, it places an order to replenish its supplies. Since the delivery time is longer, the firm must maintain a larger reserve for this contingency. Alternatively, poor infrastructure makes delivery times more uncertain and firms hold a reserve for the contingency that the delivery takes longer than average. By the transaction cost model, poor infrastructure would increase the fixed cost of each shipment, making small frequent shipments costly. This case might occur, for example, if poor infrastructure resulted in a lack of third party logistics providers who could efficiently handle small shipments. The extent of informatics technology and telecommunications development in any given country can also 7The transaction cost model predicts that large firms would hold less inventory than small firms when inventory is measured as a fraction of sales. Intuitively, a large firm can place orders in batches to capture economies of scale without spacing its orders far apart. Our analysis of firm level inventory data for several countries in Latin America did not find significant differences between inventory holdings of large and small firms. 8 affect the level and management of inventories by allowing a closer tracking of levels, demand and trends. The following simple stockout inventory model demonstrates the effect that poor infrastructure would have by increasing transport time. Assume that daily raw materials usage, which fluctuates with current or expected sales, follows a normal distribution with standard deviation 6. If daily deviations from expected usage are independent and additional inventory can be ordered immediately, the safety stock is: S= ka-oi; k = firm intolerance for running out of inventory 5 = daily standard deviation of inventory levels T = order time + transit time + handling time A firm that set k equal to 2 would run out of raw materials inventory less than 2.5% of the time.8 According to this model, if your supplier is located across the street, you don't need to hold any safety stock as long as the supplier holds finished goods inventories. On the other hand, if the supplier is located two weeks away, an unexpected increase in raw materials consumption during any two week period must be met from raw materials inventory. If daily deviations were instead perfectly correlated-more demand today means more demand tomorrow-then the safety stock would be proportional to T rather than the square root of T. Inventories should be affected by a number of other factors which are common to developing countries that we will try to control for. First, developing countries which import intermediate goods as manufacturing inputs are likely to have higher inventory levels because the import of raw materials involves longer and more uncertain delivery times as well as greater transaction costs leading to larger and less-frequent shipments. We know from our analysis of 9 data for a few Latin American countries which require frms to account separately for domestic and imported inputs that inventories of imported inputs are much higher. Second, a poorly functioning market can lead to shortages of certain goods; firms expecting these shortages would stock up on inventories in anticipation. In the Soviet Union, firms were known to maintain a high ratio of raw materials inventories to finished goods inventories for this reason.9 Third, higher uncertainty of demand should lead firms to keep higher inventories according to the stockout model. Finally, the interest rates at which firms can borrow working capital determine the holding cost of the inventories. The higher the interest rates, the costlier are inventory levels; thus one would expect lower levels in equilibrium. Since developing countries have higher interest rates than developed countries, it is then, on that account surprising that their inventories are higher. There are a number of additional factors that ideally should be included in the study but cannot due to a lack of cross-country data. If developing countries were more likely to use FIFO accounting while developed countries used LIFO accounting, their inventory stocks would appear to be higher, especially in cases of high inflation. Although we do not have evidence by country on this issue, our research on this topic indicates that LIFO, although allowed in the United States for tax purposes, is rare in both developing and other developed countries. 10 Other relevant factors are the degree of vertical integration, the concentration of upstream suppliers, production to stock vs. production to order, and the type of production technology. 1 ' We do s Since any additional orders would incur some fixed ordering costs, the firm may actually prefer a higher safety stock than indicated. 9 Chikan [1991] shows that socialist countries held a larger ratio of raw materials inventories to finished goods inventories. 10 See, for example, Nobes and Parker [1995], p.162. 11 For example, due to the fixed costs of rampup, it is more costly to run out of inventory in a continuous or batch process than in a discrete process. 10 control for some omitted variables by including GDP/capita as our measure of level of development in all of the regressions. Our approach in this paper is that high inventories are an optimal response to particular characteristics of a developing country. An alternative approach is that high inventories represent firm inefficiency, a result of poor management perhaps. We would not expect this type of inefficiency to be correlated with any of our variables once we control for level of development, and, for this reason, we do not address this type of explanation but rather focus on correlation with country characteristics. 3. Data Description It is difficult to obtain consistent time series data on inventory holdings for developing countries. The aggregate data reported in the national accounts is the change in inventories rather than the stock of inventories; often this data is based not on an inventory survey but on the difference between production and sales which leads to highly inaccurate data.12 Most national statistics agencies do have inventory stock data but they do not publish it. In order to report the size of the country's industrial production, the statistics agency typically carries out a firm survey or census, which asks about total inventory holdings at the beginning or end of the year. More detailed surveys break down inventories into three or more categories: raw materials inventory, goods-in-process inventory, and finished goods inventory. Many surveys also request data on raw materials consumed in production. The United Nations, in its World Programme of Industrial Statistics, surveyed the statistics departments of countries around the world, requesting 12 We note, however, that the initials results of our research using the aggregate inventory levels computed from the National Accounts data were not inconsistent with the stylized observation that developing countries hold more inventory than developed countries. 11 industrial data for 1973 and 1983.13 In some cases, this data was provided for an adjacent year but not the year requested. Table 2 describes the data in more detail. 31 countries provided data on inventories for the 1973 survey and 43 countries provided data on inventories for the 1983 survey, yielding a database of inventory data for 52 countries for one or two years. These data were sufficient to calculate the following end-of-year inventory levels: Raw Materials Inventory level (EOY) = Raw Materials Stock (EOY) Raw Materials Consumed Final & Process Inventory level (EOY) = Total Stock (EOY) - Raw Materials Stock (EOY) Sales Beginning-of-year inventories were also reported, permitting the calculation of another set of inventory levels. Implicit in these calculations is the assumption that inventory levels at a particular point in time are representative of average inventory levels. Since the data are for the entire industry, inventory cycles of individual firms are not important. We do not have to worry that one firm places its orders early in the month as long as another firm orders late in the month. Nevertheless, if inventory cycles are correlated between firms, then the estimate of inventory levels would be inaccurate. For example, if firms consistently run out of inventory after Christmas, then using end-of-year inventory levels would underestimate average inventory levels. Empirically, inventories, at least in developed countries, are cyclical and measuring inventory at any particular point in time may underestimate or overestimate the average inventory level of that country; a country that appears to have high inventory levels may simply be at the top of the cycle. Many developing countries have high rates of inflation leading to additional biases in the inventory level measurements. For example, under a constant annual 13 Unfortunately, this program was discontinued after 1983. 12 inflation rate of 10%, real output of $100, an inventory level of 20% and a FIFO accounting system, nominal output would be about $105 and inventory levels, as measured by the above formulas, would be 19% in the beginning of the year and 21% at the end of the year.14 We compensate for these problems in part by using both beginning-of-year and end-of-year inventories and also using two years for the same country when available. Table 2: Data Availability Country 1973 Survey 1983 Survey Country 1973 Survey 1983 Survey 1. Australia 1973 *1984 29. Luxembourg 1973 2. Austria 1973 1983 30. Macau 1983 3. Bangladesh 1982 31. Malaysia 1983 4. Barbados 1983 32. Malta 1983 5. Brazil *1973 33. Mexico 1983 6. Canada 1973 34. Netherlands 1974 7. Chile * 1973 1983 35. New Zealand 1983 8. Colombia 1973 1983 36. Norway 1973 1983 9. Costa Rica * 1980 37. Panama 1973 1981 10. Cyprus 1972 1981 38. Peru 1973 *1982 11. Czechoslovakia 1973 1983 39. Philippines 1972 1983 12. Denmark 1973 1983 40. Poland 1983 13. Ecuador 1983 41. Portugal 1971 14. Egypt 1979 42. Puerto Rico 1972 15. El Salvador 1983 43. Qatar 1983 16. Fiji 1983 44. Singapore 1973 1983 17. Finland 1983 45. Sweden 1973 1983 18. France *1983 46. Thailand 1982 19. Guatemala 1974 *1983 47. Turkey 1970 1983 20. Honduras 1975 48. UK 1973 *1983 21. HongKong 1973 1983 49. US 1972 *1982 22. Hungary 1973 1983 50. Venezuela 1984 23. Iceland 1983 51. Zambia 1973 24. Iran 1983 52. Zimbabwe 1983 25. Israel 1972 1982 26. Japan 1973 * Indicates that only total inventory data was available, 27. Korea 1973 1983 rather than both total inventory and raw materials 2. Kra17193inventory. 28. Kuwait 1974 1983 14 BOY inventory would be 20 and EOY inventory would be 22. 20/105 is about 19% and 22/105 is about 21%. 13 4. Analysis of Determinants of Raw Materials Inventory The median raw materials inventory level in our sample over all countries is .21 which means that the median industry holds enough inputs to cover two and a half months of production. For comparison, the median industry in the United States in 1972 had a raw materials inventory level of.11 representing less than one and a halfmonths of use. 10% of our dataset has raw materials levels greater than .5 and 2% has levels greater than 1. Lumpiness and volatility in commodity markets are the most likely explanations of these levels. For most of our analysis we drop any data with raw materials greater than .5 although our results do not depend on the choice of this particular cutoff.15 For final and process goods inventory, the median for the whole sample is .08 while this figure is .09 for the United States. 99% ofthis data is less than 0.35. The two sets of inventory levels are only weakly positively correlated with a correlation coefficient of 0.25. We start our analysis by regressing inventory levels on industry and country dummy variables as follows where i and c index industries and countries covered: Inventory Level,c = A /,- IndustryDummyi + z y. Country Dummy, + el,c i c The country coefficients y are graphed against GDP/capita in Figure 1. We can see that raw materials inventory is negatively correlated with GDP/capita while the relationship with final and process inventory is less clear. 15 More than half ofthe data for industry 314 (Tobacco processing) exceeded .5. Omitting industry 314 from the regressions entirely does not affect our results. The remainder of datapoints with raw materials inventory greater than .5 are broadly distributed over all industries. Egypt, Kuwait and Panama had a disproportionate share of these inventories, but excluding these countries also does not significantly affect the results. 14 Figure 1: Raw Materials Inventory vs. GDP/Capita 379486 Egypt Zimbabwe Peru _Hcncbjres El Saoho Raw Guntemnl Materials Bmtglade pwFs Inventory _nnaad aow (fraction of Phlppi VW a year) Luwbou Col4ibib FPi CwwSb _ urk_poloysin AWR knord PuLto R U10$5CW o) HorigKon ism ~~Us .103652 Koe 6.93803 9.59106 Log of Real PPP GDP/apl Final and Process Inventory vs. GDP/capita .169216 - POaUlS Sweden Egypt Cotaft Na Ecuador Fhbn B M FYI knUk 7imbabwN Peru Final and Cypnu AWhStt Process PhilpWisian Twkey brw Dewsima Inventory - Uondura Us (fraction of a El Salv jombia Mal New ZeI Cvnda year) mJalvf3w Sk9wpa Chile AsM Korea Vlnx _Snezil Hwgwy Puguo R Bedwwos Lwnmbou Japan Potand HongKcn .054599 l klnd 6.93803 Log of Real PPP GDPkaplta (aere 197i164) 9.59106 15 For the remainder of this section, we focus on raw materials inventory. We replace the country coefficients in the regression with country characteristics; essentially, we are trying to explain these coefficients using country characteristics. Inventory Level,c = E * IndustryDummyi + X A, . Country Characteristicx + ,ec All reported standard errors in the regressions are robust standard errors corrected for clustering at the country level. Resolving which particular characteristic of developing countries leads to high inventory levels is made difficult by the fact that we are starting with a dataset of only 52 countries. The independent variables that we are interested in are not available for all countries and some variables, such as infrastructure and GDP/capita, are highly correlated with each other, making it difficult to differentiate between explanatory variables. Nevertheless, we do obtain significant results in our regressions. Table 3 describes the variables and Table 4 summarizes the values they can take. 16 Table 3: Description of Explanatory Variables Telephone mainlines per "Telephone mainlines are telephone lines connecting a customer's equipment to the public person switched telephone network." Data are the averages of available years over the period 1971- 1985. Infrastructure Quality Assessment of the "facilities" for and ease of communication between headquarters and the operation, and within the country," as well as the quality of the transportation. Average data for the years 1972 to 1995. Scale from 0 to 10 with higher scores for superior quality. Source: BERI's Operation Risk Index as used in La Porta et al [19991. Transfers and subsidies/GDP Total govermnent transfers and subsidies as a percentage of expenditure multiplied by government consumption as a percentage of GDP. "Subsidies and other current transfers include all unrequited, nonrepayable transfers on current account to private and public enterprises, and the cost of covering the cash operating deficits of departmental enterprise sales to the public. Data are shown for central govermment only. General government consumption includes all current expenditures for purchases of goods and services by all levels of government, excluding most government enterprises. It also includes capital expenditure on national defense and security." Data are the average of available years over the period 1971-1985. Log GDP per capita Logarithm of PPP GDP per capita measured in 1985 dollars. Data are the averages over the period 1971-1985. Source: Penn World Tables (Mark 5.6). GDP Growth "Annual percentage growth rate of GDP at market prices based on constant local currency." Data are the averages for available years over the period 1971-1985. Lending Interest Rate "Lending interest rate is the rate charged by banks on loans to prime customers." Real lending rate is computed using GDP deflator. Data are the average over all available years in the period 1971-1985. Imports/GDP "Imports of goods and services represent the value of all goods and other market services provided to the world. Included is the value of merchandise, freight, insurance, travel, and other nonfactor services. Factor and property income (formerly called factor services), such as investment income, interest, and labor income, is excluded." Data are the averages for available years over the period 1971-1985. Data source for explanatory variables is the 1999 World Development Indicators on CD-ROM unless otherwise noted. 17 Table 4: Summary of Explanatory Variables Variable Countries Mean Std. Dev. Min Max Log GDP/capita 48 8.46 0.75 6.93 9.61 Telephone mainlines per person 52 0.16 0.15 0.00 0.57 Infrastructure Quality 31 6.05 1.84 2.50 9.15 Transfers & Subsidies/GDP 46 0.06 0.04 0.00 0.18 hnports/GDP 49 0.38 0.29 0.09 1.74 Exports/GDP 50 0.37 0.29 0.04 1.65 Lending interest rate (real) 36 0.08 0.15 -0.17 0.81 Growth - Level 49 3.74 2.16 -1.37 8.14 Growth - Standard deviation 48 4.11 2.01 1.54 10.86 Regressions (1), (2), and (3) of Table 5 display the results of regressing raw materials inventory on infrastructure and the presence of a free market, as well as some control variables. We use two proxies for infrastructure, telephone mainlines per person and BERI's infrastructure quality index, which, although more comprehensive, is available for fewer countries. These proxies for infrastructure are significant at the 1% or 5% level; the coefficients suggest that a one-standard deviation worsening in infrastructure increases inventories by 27% to 47% relative to U.S. levels. 16 Our proxy for the lack of a free market is transfers and subsidies to private and public enterprises expressed as a fraction of GDP. 17 A one-standard deviation restriction on the free market increases raw materials inventories by 19% to 30%. 16 As shown in Figure 1, the U.S. has one of the lowest levels of raw materials inventory. Comparisons with other countries as the denominator would produce a smaller percentage effect. 17 In another version of this paper, we used stated-owned enterprises and business regulation as two alternate proxies and obtained significant but smaller effects. 18 Table 5: Regressions (1) (2) (3) (4) (5) (6) (7) (8) .. Dependent Variable Raw Raw Raw Upstream Upstream Upstream Raw as % of Raw as % of Materials Materials Materials Inventories Inventories Inventories Raw + Raw + Upstream Upstream Log real PPP GDP/capita -0.0229 0.0010 -0.0304* -0.0328*** -0.0193* -0.0320*** 0.0444* 0.0523* (0.0186) (0.0285) (0.0171) (0..0950) (0.0103) (0.0077) (0.0227) (0.0274) Telephone mainlines per -0.2934*** -0.1968** 0.0950* -0.0926 -0.5417*** person (0.0948) (0.0928) (0.0539) (0.0549) (0.1695) Infrastructure Quality -0.0300*** 0.0021 -0.0374*** (0.0086) (0.0044) (0.0076 Transfers & Subsidies/ 0.7427*** 0.4105** 0.6453** 0.2136* 0.3098** 0.6608*** 0.4385 -0.3475 GDP (0.2226) (0.1947) (0.3128) (0.1202) (0.1235) (0.1238) (0.4809) (0.4063) Imports/GDP 0.0290* 0.0372*** 0.0449 -0.1765 0.1615 (0.0166) (0.0124) (0.0296) (.1598) (0.1596 Exports/GDP -0.0157 -0.0151 0.0158 0.2721 -0.0767 (0.0108) (0.0111) (0.0110) (0.1798) (0.1856) Lending Interest rate -0.0317 -0.0442*** (real) (0.0368) (0.0149) GDP Growth -0.0113 -0.0038** (0.0073) (0.0016) GDP Growth Standard 0.0108 -0.0066** Deviation (0.0075) (0.0019) (24 industry dummy Included Included Included Included Included Included Included Included variables) I__ I_I I # of clusters (countries) 42 29 31 44 30 32 41 29 R-Squared .2528 .2897 .2846 .3893 .4291 .4549 .3234 .35 18 # of Observations 2086 1627 1408 1962 1642 1271 1554 1307 Robust standard errors corrected for clustering at the country level are in parenthesis. *Indicated significance at the 10%/o level; **Indicates significant at the 5% level; ***Indicates significant at the 1% level. Inventories greater than .5 have been dropped for these regressions. Coefficients in regressions (1)-(6) represent the effect of an absolute change in the explanatory variable on inventory level expressed as fraction of a year. For example, if telephone mainlines per person increased from .5 to .6 in regression (1), inventories would fall by .02934 of a year or about 11 days. Coefficients in regression (7) and (8) represent the effect of an absolute change in the explanatory variable on the percentage of inventories held as raw materials. For example, if telephone mainlines increased from .5 to .6 in regression (7), 5.417% of inventories more are held as raw materials. In the U.S., the median industry holds 57% of inventories as raw materials so that the . I change in telephone mainlines leads to a 9% change in holdings. 19 Table 6: Input-Output Table 311 314 321 322 323 331 332 341 342 351 352 353 355 356 362 369 371 372 381 382 383 384 385 390 311 29%/ 0% 0% 0% 200%o 0% 0%/o 1% 0%/- 1% 3% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% ° 0%N 314 0% 34% 0%/ 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%/ 0% 0% 0% 0% 0% 0% 0% 0% 0% 0%/ 0%/ 321 0% 0% 43% 48% 9% 0% 20% 1% 0%/ 0% 0% 0%/ 5% 0% 0% 1% 0% 0% 0% 0%0% 3% 2% 4%T 322 0% 0% 1% 41% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 0% 1% 323 0% 0%1 1% 1% 51%. 0% 0% 0% 0%0 - 0% 0% 0% 0%° 0% 0% 0% 0% 0% 0% 0% - 0%° 0% 0% - 0%N 331 0% 0% 00% 0% 0% 66% 22% 12 0% 0% 0% 0% 0% 0% 4% 2% 1% 1% 0% 1% 0% 0% 0% 5°% 332 0%/ 0%0- 0% 0%/ 0%/ 0% 1% 0%/o 0%0% 0% 0 0% 0% 0% 0% - 0% 0% 0 - 0% 2% 0% 0% 341 6% 9% 0% 0% 2% 1% 4% 55% 47% 2% 8% 0% 5% 4%1 14% 5%° 0% 0%1 2% 1% 2% 0% 4% 8% 342 0%/ 3% 0% 0% 0°/ °% 0 / °'0% 36% 0°/ 2% 0% 0% 0% 0% 0% 0% 0%° 0% 0%0%0% 0%0% 351 1% 0% 3% 0% 7% 1% 0% 9% 6% 60% 25% 2% 12% 74% 13% 8% 3% 1% 2% 0% 1% 1% 1% 3% 352 1 % 1% 0% I1% 1%° 0% 2% 1%/ 0% 1% 30% 0% 0%/6 1%/ 1% 1% 0% 0% 2% 0% 0% 1% 0% 2% 353 1% 1% 1% 0%/ 1%1 1% 1% 1% 1% 4% 3% 14 I% 1 1% 2% 3% 1% 1% 1% 1% 0% 0% 0% 1%° 355 4% 4% 2% 1% 6%1 2% 11% 6% 3% 3% 13% 0% 15% 9% 8% 3% 1% 2% 4% 6% 6% 7% 7% 12% 356 0%/ 5% 31% 2% 0%1 1% 1% 5% 0% 2% 6% 0% 47% 6%/a 0% 2% 0% 3% 1% 0%Wo1% 0% 2% 8°N 362 2% 0% I% 0% 0%° 1% 1% 0% 0% 0% 1% 0% 2% 0%/ 36%/ 1%j 0% 1% 0% 0% 1% 1% 2% 0°% 369 0%0% 0% 0% 0% 1% 1% 0% 0%0% OY 1% 0% 0%0% 4% 39°% 4% 1% 0% l% 1% 0% 0% 1% 371 0%0% 0% 0% 0% 0% 10 0o% 0%° 1% 0% 0% 1% 0% 0% 3Y 51% 2% 41% 24% 4% 5% 3% 6% 372 0% 0% 0% 0% 0%° 0%D 3 % 00% 0%° 0% 0% 0% 0% 0%° 00% 1 % 5% 70% 21% 9%/ 7%1 5% 4% 14% 381 5% 1% 0% 0% 1%o 5% 18% 1%oM 0% 3_ 4% 0% 2% 0% 0% 3% 6% 2 19% 14% 8%13% I 11% 6% 382 0% 0% I 1 % % 00%1 I % 2% 2% 1 % 2%1 0% 0%/ 4% 1 % 5% 2%1 9% 5%° 5% 34% 3% 7% 2% 4% 383 0%/o I % 0%M 0% 0%o 1%1 0 o/ 0 0% 0% 0% 0% 0%° 0% 0% 1 % 0% 3% 1 '/