INVESTMENT UNDER UNCERTAINTY AND FINANCIAL CRISIS -A COMPARISON OF ADJUSTMENT COST IN FOREIGN AND DOMESTIC FIRMS IN TURKEY Camilla

Jensen
Kadir Has University, Istanbul, Turkey Abstract
The objective of the paper is to test the stability hypothesis that foreign investors are relatively more insulated from uncertainty and how it spills over on their investment adjustment cost. The Q model (implying that investments are explained by the fundamental value of the firm) is implemented with reasonable success for firm level panels in Turkey. Robustness of the results and despite the general obstacle that inflation poses on the study is increased by applying different datasets with different time horizons, different measures of investment and profitability and different problems of attrition. The general finding of the study is that the stability hypothesis is confirmed. The difference in adjustment cost across domestic and foreign owned firms is particularly affected by uncertainty measures whereas the general adjustment cost difference is estimated to be small. In periods of high uncertainty it is found that the decline in the growth of the investment rate for domestic firms is at least twice as high compared to the decline in the growth of the investment rate among foreign held firms.

Keywords Foreign direct investment, private investment, uncertainty, Q model, firm-level panel data.

Introduction One of the important issues surrounding the participation of foreign capital in emerging market economies is how it affects the stability of the economic environment. From the viewpoint of emerging market populations this is an important welfare question. Financial crisis situations followed by local devaluations will affect disproportionately the majority of populations in these economies, i.e. those who are not foreign currency or net asset holders abroad and therefore especially the poor. In this perspective foreign direct investment can play an important stabilizing role, even though this may also depend on the particular circumstances, origins and scope of the crisis. The research question is whether foreign firms investment adjustment cost differ from that of domestic investors in periods of high and low uncertainty? Adjustment cost (the cost of getting from the present level of investment to that which fully realizes the fundamental value of the firm) may be generally higher in periods of high uncertainty which is a complementary explanation to the option value of waiting effect on investment with uncertainty (Dixit and Pindyck, 1994). Both phenomena may explain the tendency of investment to fall more than proportionate with the decline in output during episodes of high uncertainty. However, firms may be heterogenous in their adjustment cost function due to differences in their resource base, their external networks, corporate governance systems and market reach. Past research has shown that such heterogeneity in particular is to be detected across different basic ownership classes in emerging economies such as those related to the publicly, private domestically and foreign held classes of firms (see e.g. Harrison et al., 2004). Experiences from the Asian episodes of the late 1990s suggest that in periods of high uncertainty and financial crisis foreign firms may be acting counter-cyclical in terms of their investment decisions (Frankel and Rose, 1996, Moreno et al., 1998, Krugman, 2000, Ni and Ratti, 2009). One reason is that foreign firms may depend less on the domestic market for their resources and sales. Foreign firms may also be more likely to be decoupled from local bandwagon effects. In this case the adjustment cost of foreign investors should be less sensitive to uncertainty and especially if the source of uncertainty is largely local in character such as uncertainty that arises from values denominated in local currencies. Against the hypothesis of stability or countercyclical behavior of foreign investors is that foreign investors and especially in emerging market type of economies may be less informed relative to the domestic firms about the sources and directions of local uncertainty. This would oppositely suggest that FDI could be pro- rather than countercyclical (Ni and Ratti, 2009). Furthermore, the 2008 financial crisis has proven that trade policy nowadays also involves resource allocation decisions within multinational firms. In a scenario of general rather than local financial crisis multinationals may be prone to shift resources to the advantage of home country workers if the home country government resorts to protectionism by giving subsidies to firms. The objective of the paper is to test the stability hypothesis using available firm level panel datasets for Turkey. This is done by implementing the Q model with adjustment cost on different firm-level panels and using different measures of uncertainty. The findings confirm the stability hypothesis in Turkey during the period 1993-2007. Across all specifications the adjustment cost of domestically held firms are at least as high or higher than the adjustment cost with foreign and publicly held firms. Expanding the Q model with uncertainty as a direct adjustment cost shows that the higher cost in domestic firms may be explained by uncertainty. Results across different measures of uncertainty suggest that financial and political uncertainty causes higher adjustment cost, whereas economic uncertainty is not found to affect the cost with domestic firms or have less but a small impact on adjustment cost with foreign held firms. Section 2 gives a brief review of the empirical literature on investment under uncertainty. The data and methodology is introduced in Section 3. Section 4 explains the statistical results followed by a discussion of the results and methodology in Section 5. Literature review Investment under uncertainty has been studied in a variety of ways in the literature. Less attention has been paid to the specific role of foreign direct investment under the general heading of investment and uncertainty (see e.g. Bloningen, 2005). The issue first attracted attention during the Asian financial crisis where economies that relied more on foreign direct investment for their foreign capital flows appeared to suffer less from the crisis. Traditionally FDI has been seen as a crowding in crowding out dilemma at the macroeconomic level vis--vis domestic investment (Agosin and Machado, 2005). With the availability of firm level panels there is the possibility now to rethink many of these research questions. Since the focus in the present study is firm-level decisions about investment I mainly focus on how uncertainty has been found to affect investment decisions at the firm level in the literature. Aggregate studies of investment and uncertainty in developing countries have generally confirmed the prevailing expectation that uncertainty will reduce investment (Serven, 1998). At the microeconomic or firm-level there has been stronger hesitation to model uncertainty because adequate measures can be difficult to find under assumptions of heterogenous firms with subjective

uncertainty (Bond and Reenen, 2007). According to the review in Bond and Reenen (2007) the main mechanism through which uncertainty affects investment decisions within the Q model framework is through a lower transmission mechanism from short run fluctuations in demand to investment under episodes of high uncertainty. This has been documented for manufacturing firms in the UK and Italy. A disturbing factor for the classical Q model in this context is that the stock market measure of q most likely will be correlated with the exogenous measures of uncertainty. Overlapping with the question about uncertainty and how it affects the adjustment cost is the issue of financial constraints on firms abilities to invest. Harrison et al. (2004) show in a multi country study that multinational firms are generally less constrained and may also have a positive spillover effect on domestic firms by relieving their finance constraints. For the Ivory Coast, Harrison and McMillan (2003) found that domestic private firms are more financially constrained relative to foreign firms whereas publicly held firms are not constrained. Gezici (2007) studied the same question using a firm-level panel on Turkey and also investigated for the impact of uncertainty on investment with Turkish firms. Generally financial liberalization was found not be associated with a relief of the financial constraint whereas there was found a strong negative causation running from uncertainty to investment in the firm-level panel for Turkey. Gezici (2007) did not focus on ownership, but showed that export intensive firms in Turkey are less affected by uncertainty in their investment decisions. Data and methodology The Q model The Q model is described in detail in Bond and Reenen (2007) and in Bond and Cummins (2001) and in Blundell et al. (1992). The assumptions of the wealth maximizing firm collapses into an easily implementable functional form using that Tobins q is an approximation to the value of the firm. In other words q can be used as an indication that there are unexhausted investment possibilities existing within firms. For the ith firm the rate of investment (I/K) is then described as a constant c (investment for which there is no adjustment cost), and a linearly increasing function on average q, slope of which depends on the size of adjustment cost b.

It 1 1 $ c # !qt % 1" # et $ c # Qt # et Kt b b

(1)

The larger the adjustment cost b the slower the firm will be at realizing its full potential. When all investment possibilities have been exhausted the (marginal) q value should approach 1 e.g. the marginal value of the last unit of investment equals its opportunity cost (q=1 or Q=0). The lack of immediate adjustment to the optimal capital stock in the Q model is fundamentally explained by an adjustment cost function which is assumed to be convex in the level of investment. The exact shape of the adjustment cost function may depend on a host of factors such as institutions, government regulation and taxation, financial constraints and last but not least uncertainty facing firms in their operating environment. However, the general assumption is that firms operating in the same environment face the same adjustment cost function. This restriction I attempt to lift by estimating whether different firms such as multinationals and domestic firms operating in Turkey may have different adjustment cost functions. The advantage of the Q model compared to other structural models is that it is relatively simple to implement and has lower data requirements. As described extensively in a handbook chapter by Bond and Reenen (2007), relative to other reduced form type of models it involves the estimation of deep or real parameters that can be related back to the theory of the profit maximizing (neoclassical) firm. One disadvantage is that it involves standard assumptions such as constant returns to scale technology and competitive markets in order to use average q (being an observable either using stock prices or expected profits i.e. the average future expected value of the invested capital) rather than marginal q (a non-observable e.g. the future expected value of the last unit of investment). If firms exhibit economies of scale or market power the adjustment cost may no longer be convex. Adjustment costs themselves may also vary with different types of investment e.g. investment in intangible assets such as R&D can give rise to dynamic scale economies. However, from a short run adjustment perspective with exhausted static scale economies and sticky prices the assumption of convex adjustment cost may not be unreasonable for other market forms than perfect competition. Since many of the firms in the samples used are not listed on the stock exchange I follow Bond and Cummins (2001) by using an alternative measure to the classical Tobins q that has been applied in most previous studies using the Q model. How the fundamental value of the firm is measured using published Turkish accounting data is described in the next section.

Description of the data Public records or censuses giving firm level information are only scarcely available in the case of Turkey. This is true in particular for the historical perspective. High quality investment data is only available for countries where such data has been collected with that particular objective. The present study relies on secondary data sources from the Orbis database published by Bureau Van Dijk in Holland together with a longer panel provided by ISO (the Chamber of Commerce in Istanbul) (in the present paper top 500 is used). Both datasets suffer from lack of direct ability to observe annual investments incurred by firms. The two datasets are different on a number of factors which is both a weakness and strength. The main difficulty lies in the measurement of the investment variable. Due to a relatively weak overlap of the firm populations in the two panels it was chosen to treat them as complementary but independent datasets. The ISO dataset is of higher quality in view to the persistency of the data variables. But a weakness of the panel is attrition since firms that drop out of the top 500 cannot be followed and will only return into the dataset when their performance improves. How this kind of attrition affects the results is presently unknown. The advantage of the Orbis panel is that it does not share this feature since it persistently follows firms for a ten year period as long as the firms remain active producers. The disadvantage is a lower persistency for individual data variables across the observed units. For example, information about ownership is not persistently available across the whole sample. Also the Orbis dataset has missing information for important economic variables such as the number of workers. A major difference between the two panels concerns the measurement of investment. Only a full measure of investment is available from Orbis including all fixed assets (and separable by tangibles and intangibles) no matter their type of financing. For the ISO data there is the possibility only to observe investments when financed internally by the firm either through equity capital or through the owners private capital or both. Despite the reduced ability to observe investments with the ISO data there is the advantage that the financing constraint of different firms is assumed away because in principle all the externally raised capital has been reduced out both of the nominator (EBT instead of EBIT) and the denominator (net assets instead of total assets) when estimating the ROA ratio. However, this gives rise to an unobserved heterogeneity in the net ROA (Panel A ROA) of firms because the external financing may give an unexplained return effect over and above the cost of capital. The repercussions hereof are smaller in a panel because such firm-specific effects can be estimated as fixed effects or reduced out using the GMM estimator. The exact definition of variables across the two datasets is explained with Table 1. Descriptive statistics for both panels are given in the Appendix which also shows the continuous and discrete values of the risk variables. Figure 1 gives an overview of the aggregate trend in private and public investment in Turkey together with estimates of foreign investment through annual FDI and portfolio inflows as published by the World Bank. The data are only indicative and especially for FDI inflows with respect to the foreign held shares of capital in Turkey. For example, foreign held firms may finance their investment through host country sources that are not registered with the balance of payments such as retained earnings and through local banks. Turkey has during the period of study suffered from a minor crisis in 1998 and a major crisis in 2001. Whereas the first crisis was international in character the second crisis was of a more internal character. Among other did Turkey suffer from an earthquake in 1999 that affected several parts of Istanbul. In the aftermath followed major drops in local values especially real estate including significant strains on the public finance situation.

FIGURE 1: Investment in Turkey by sector (left axis) or source (right axis), 1980-2007 (in mio. YTL, 1987-prices)

800 600 FDI inflow 400 200 0 Portfolio inflow 30,000 20,000 10,000 Public 0 1980
Source:

-200 -400

Private

1985

1990

1995

2000

2005

Turkstat (2008): Historical Statistical Indicators, Turkish Statistical Institute, Ankara and WB (2009): World Development Indicators. The World Bank, Washington D.C.

Table 1: Definition of variables in the firm-level panels Variable I/Kit Panel A (ISO) The investment rate is approx. by using the within year change in real net assets to the real net assets at the beginning of the year. It is assumed that all assets have been rewritten at current cost. Not available in Panel A. Real profits are estimated in their net Form by dividing earnings before taxes with net assets. It is assumed that net assets have been rewritten at their current cost in the accounts. Real physical capital is approximated with net assets and using the wholesale price index published by the Central Bank. qit is estimated by dividing ROAit with rt. Domestically privately held firms are classified as all firms that are 100% owned by Turkish nationals. Foreign owned firms are classified as all firms with a significant share of (>10%) foreign capital participation. Publically owned firms are classified as all firms with a significant (>10%) public capital participation. Panel B (Orbis) The investment rate is approx. by using the within year change in real fixed assets to the real fixed assets a the beginning of the year. It is assumed that all assets have been rewritten at current cost. Real intangible assets deflated using the wholesale price index published by the Central Bank. Real profits are estimated in their gross form by dividing EBIT with total assets. It is assumed that total assets have been rewritten at their current cost in the accounts. Real physical capital is approximated with fixed assets using the wholesale price index published by the Central Bank. qit is estimated by dividing ROAit with rt. Domestically held firms are classified as firms where the global ultimate owner is Turkish. The ultimate owner is defined as holding more than 25% of the total assets. Foreign owned firms are classified as firms where the global ultimate owner is foreign. The ultimate owner is defined as holding more than 25% of assets. Not available in Panel B.

IA ROAit

Kit qit DOMi

FORi

PUBi

General variables rt RISKEt RISKFt RISKPt The real interest rate or the opportunity cost of investing in the firm. It is set objectively (risk free) at 5% in the study. Economic risk or uncertainty, estimated as a continuous variable using the annual SD in the monthly wholesale price index. As a discrete variable it is classified as high (1) when it is unusually high and otherwise as low (0). See also Appendix Table A1. Financial risk or uncertainty, estimated as a continuous variable using the annual SD in the daily percentage changes of the USD/TL exchange rate. As a discrete variable it is classified as high (1) when unusually high and otherwise as low (0). See also Appendix Table A1. Political risk or uncertainty, estimated with the civil rights index published by Freedom House. The variable is discrete and is classified as high (1) when civil rights are registered by Freedom House to be diminishing and otherwise low (0) combined with years where the political system in Turkey was subject to unusual events. See also App. Table A1.

Following Bond and Cummins (2001) q is approximated by using information about profitability. However, since estimates of expected profits are not available for Turkey, reported profits are used instead. This reduces the futuristic dimension in the estimate of the fundamental value of firms. But forecasts will typically be based strongly on reports of actual profits. Furthermore, using forecasts may introduce a reverse causality problem since signaling through investment without realized potential may lead brokers to revise upwards the expected profitability in which case the estimated q is a function of I/K in (1) rather than vice versa. q is approximated as follows:

EBIT EBIT ROA r q$ & K & r r Total Assets

(2)

According to the results of Bond and Cummins (2001) this approximation or alternative accounting measure to the classical Tobins q reduces the error introduced in the Q model due to stock market bust and booms. Bond and Cummins show that a stockmarket based measure may introduce not only extra white noise but significant econometric problems (simultaneity bias) due to the correlation between the error term and q on a cross sectional basis in firm-level panels. The choice of real interest rate should reflect a similar investment opportunity with the same risk profile. To find such an interest rate in Turkey during the period covered is a difficult question. Macroeconomic experts suggest using government bond yields which gives the truest estimate to a real exchange rate in the economic environment prevailing in Turkey in the 1990s. However, its levels reflect high uncertainty driven by fiscal misconduct or lack of Central Bank independence during the 1990s up until after the financial crisis. Oppositely if the return on more mundane savings instruments such as bank deposits is chosen the real rate of interest is negative throughout the 1990s (most likely because they are crowded out by the high returns on government bonds). Hence the real rate of interest is set at 5% e.g. a realistic return obtainable in a western bank at low risk. By this choice the influence of uncertainty on firm agency, e.g. that which reflect the psychology of managers but not actual sales levels, profits etc. (since it is reflected in Q) are shifted to be observed as part of the adjustment cost function. The derivation of the investment rate (or growth rate) and the profitability rate (ROA) is based on the assumption that international accounting standards have been adopted by Turkish firms throughout the period of study. According to unpublished Turkish sources this may in fact only be true for all firms after 2004 when the standards became universally enforced. Firms who do not follow these standards may have been screened out by Bureau van Dijk in Panel B and in Panel A also ISO will help to drive standards among Turkeys largest, strongly internationalized and highly professional firms. Only in Panel B was it necessary to remove extreme outliers (most likely caused by miscalculations or misleading information about inflation and/or exchange rates) e.g. firms with positive or negative investment rates in excess of 10 (1,000 percentage). Finally it should be noted that ownership is measured more precisely in Panel B because ISO observes every year the ownership situation with firms. In the Orbis data only the most recent ownership information will be attached to the financial accounts. Implementing the Q model with the available data The Q model is implemented using the General Methods of Moments estimator. This is standard in the literature and the main advantage of this particular estimator is that cross section effects are differenced out. Another major advantage is that the method is robust to cross-sectional heteroscedasticity in the error term that may remain after the fixed effects have been removed. Main challenges are that long panels are required for the method to be robust and it may be difficult to identify appropriate instruments. Standard requirements to the instruments used are that they must be correlated with the included endogenous variables but independent of the unobservable error process that creates heteroscedasticity. Different instruments are tried out and generally it is found that the lagged dependent variable (2 periods) is not an appropriate instrument. Instead is used sales, intangible assets and/or labour as instruments all lagged two periods. The equation estimated looks as follows in the basic model (note that standard controls such as firm-specific effects, industry, age, region etc. - are all structural factors and hence have been reduced out):

I I $* + # ) +Qit # (t # ' it K it K it %1

(3)

The fact that there may remain heteroscedasticity in this equation after first differencing is also due to a fixed or unobservable structural factor that firms respond in heterogenous but consistent ways over time in their response to profitability. Subsequently, and removing the time effects because they are directly correlated with the risk dummies, is estimated the Q model with uncertainty modeled as a factor that directly affects the adjustment cost of firms:

I I $* + # ) +Qit # , E +Qit RISKEt # , F +Qit RISKFt # , P +Qit RISKPt # ' it K it K it %1

(4)

Statistical results The results of implementing the model on the data are shown in Tables 2 and 3. The adjustment cost are found to be generally higher in Panel A (a 1 percent increase in the value of the firm gives rise to an 0.02 percent increase in the investment rate) than Panel B (the investment elasticity increases to 0.06). Compared to the adjustment cost found by Bond and Cummins (2001) using comparable data, Turkish firms face more barriers to investment. One reason may be that adjustment cost in emerging economies are higher especially because of uncertainty or risk factors which will tend to drive up the real interest rate and therefore crowd out investments with firms. Another reason is that if Q in the present study is overestimated due to the inflationary problems of valuation in the firm accounts then the adjustment cost may be overestimated. Furthermore, adjustment cost in Panel A relative to Panel B are overstated because there is only the focus on internally financed (through owners own or equity capital) investments. The stability hypothesis is confirmed following the results in Table 2. The adjustment cost of foreign firms are estimated to be half of those prevailing with domestic firms. Whereas publicly held firms are found to have prohibitively high adjustment cost. However, the results for this minor part of the sample are less reliable due to the low number of observations available (e.g. the Sargan test of the instruments does not pass even at the 50% probability level). Similarly, the results for ownership differences are less valid in Panel B due to lower sample size and especially lower panel length. Note that the GMM estimator requires that at least the first two years of observations are sacrificed to the model. Due to lower sample size there is less opportunity to observe investments with foreign held firms. (According to the more representative Top 500 data from ISO foreign firms average capital contribution to Turkish manufacturing was around 10% in 1993 and rose to 20% in 2007). Therefore the Q model can only be estimated for firms that are not foreign held (hence domestically or publically held). Despite the inaccuracy of the information in Panel B the adjustment cost are confirmed to be higher for the purely domestically held firms compared to the same cost in the average firm in the full sample. In Table 2 the result of implementing Equation 4 on the data are shown. In most cases the model test statistics improve by removing the period specific effects and replacing them with adjustment cost that vary with the level of uncertainty prevailing in the three dimensions of economy, finance and politics. Introducing uncertainty leads to an estimate of generally lower level of adjustment cost in periods of low uncertainty. Economic risk factors are found not to place a significant burden on firms. Only foreign firms are found to suffer from economic risk. This result might be explained by the fact that Turkish firms are used to operating in a high inflation environment, whereas foreign firms in Turkey typically come from a low inflation environment. Financial and political uncertainty is found to be generally detrimental for investment with firms in Turkey. All firms suffer more under these types of uncertainty, however, foreign firms suffer much less from these sources of uncertainty whereas publicly held firms appear to be immune to uncertainty in general (which also needs to be interpreted against their generally very high level of adjustment cost and low sample size which reduces the reliability of those particular results). In Panel B the difference in adjustment cost elasticity across all firms and domestic firms is around 2 whereas the difference in Panel B which directly enables the ability to compare across specific ownership classes the difference is estimated to be as high as 5. Results for political uncertainty are very similar although the adjustment cost are affected slightly less by political uncertainty relative to financial uncertainty.

TABLE 2: BASIC Q MODEL (DEP. VAR. IS !I/K IN FIRST DIFFERENCE) t-statistics are reported in parenthesis All firms LDV (!IK(-1)) !Q Year FEs N (obs) SE of reg. J-stat I-rank Sargan-test -0.070** (-1.91) 0.020 *** (13.28) Yes*** 3,695 0.41 191 192 0.24 PANEL A Domestic -0.174*** (-24.22) 0.020*** (12.84) Yes*** 2,438 0.39 212 192 0.04 Foreign -0.116*** (-209.52) 0.037*** (196.72) Yes*** 820 0.32 158 169 0.42 Public -0.130*** (-11.96) 0.009*** (31.47) Yes*** 242 0.41 44.6 55 0.84 PANEL B All firms -0.008*** (-8.04) 0.058 *** (99.62) Yes*** 440 0.36 70.9 76 0.35 Domestic 0.011*** (32.01) 0.047*** (1406.06) Yes*** 403 0.36 71.2 74 0.28

TABLE 3: THE Q MODEL WITH UNCERTAINTY (DEP. VAR. IS !I/K IN FIRST DIFFERENCE) t-statistics are reported in parenthesis PANEL A All firms Domestic Foreign Public LDV (!IK(-1)) -0.047*** -0.250*** -0.172*** -0.253*** (-7.39) (-61.15) (-968.03) (-682.19) !Q 0.037*** 0.052*** 0.049*** 0.011*** (20.50) (16.47) (3389.19) (345.34) !Q*RISKE 0.033*** 0.020*** -0.023*** 0.009*** (10.27) (5.32) (-239.24) (8.75) !Q*RISKF -0.016*** -0.048*** -0.009*** 0.002** (-7.67) (-9.60) (-84.33) (2.32) !Q*RISKP -0.015*** -0.038*** -0.018*** -0.000** (-9.78) (-12.04) (-758.19) (-2.66) N (obs) 3,695 2,438 820 242 SE of reg. 0.44 0.41 0.33 0.41 J-stat 243 213 159.3 50.13 I-rank 180 180 165 54 Sargan-test 0.000 0.03 0.50 0.43

PANEL B All firms -0.047*** (-1154.02) 0.070*** (3165.53) 0.092*** (505.17) -0.038*** (-1565.30) -0.027*** (-1238.53) 440 0.37 68.1 69 0.34

Domestic -0.029*** (-8.04) 0.0589*** (1154.86) 0.115*** (258.28) -0.078*** (-1491.65) -0.072*** (-396.60) 403 0.37 64.5 66 0.36

10

Discussion Since the quantitative results have been commented on in the previous and introductory sections I will in the concluding discussion focus on issues that are relevant for the overall methodological framework applied in the paper towards investigating the relationship between investment and uncertainty. Research on the topic of investment and uncertainty using firm level panels is only in its infancy as described in the survey by Bond and Reenen (2007). Their preliminary survey on the relationship suggests that it is the responsiveness of investment with respect to demand or sales at the firm level that is affected by uncertainty. In this paper I have avoided to explore this avenue because if sales is used a control factor in the Q model for e.g. firm size it is one of the factors that reduces out using the GMM estimator. Besides this there is no theoretical justification for including sales in the model and there is the additional problem that it is quite highly correlated with the capital variable (especially when measured appropriately) which may be one important reason why its inclusion tends to crowd out the Q factor in studies of this type (simply because K enters in the denominator of Tobins q or ROA). Also, it should not be sales but profitability that is the barometer of the long run viability of investing with firms. In this paper I also tried to avoid other standard problems of the Q model that could prevent to investigate for the relationship with uncertainty. For example I avoided using Tobins q directly since it will be affected strongly by episodes of uncertainty. I also decided in the end not to use the real rate of interest when calculating the fundamental value of the firm since the real rate of interest may be strongly correlated with the measures of uncertainty that I use. One major problem with the way I have implemented the Q model may be the unexplored nature of the relationship between investment, uncertainty and financial constraints. Uncertainty and financial constraints are often dealt with separately in the literature but it is clear that they are closely related and arise out of the same institutional problems of capital market imperfections. Generally using two different panels increases the reliability of the obtained results, because the results are critically assessed using triangulation as a tool to this end. This is in particular important because of the fragility of the investment rate in Panel A, where there is only the possibility to observe new investments arising out of internal assets (net assets). Even though Panel B has other problems, the availability here of more sound measures of investment based on the assessment of fixed assets greatly increases the reliability of the overall results.

11

Acknowledgements The research has benefited from a Kadir Has University research grant. I am grateful to Istanbul Sanaiye Odasi for providing the data. References Agosin, Manuel R. and Roberto Machado (2005): Foreign Investment in Developing Countries: Does it Crowd in Domestic Investment?, Oxford Development Studies, Vol. 33, no. 2, pp 149-162. Bloningen, Bruce A. (2005): A Review of the Empirical Literature on FDI Determinants, Atlantic Economic Journal, Vol. 33, No. 4. Blundell, Richard, Stephen Bond, Michael Devereux and Fabio Schiantarelli (1992): Investment and Tobins Q, Journal of Econometrics, Vol. 51, pp 233-257. Bond, Stephen R. And Jason G. Cummins (2001): Noisy Share Prices and the Q Model of Investment, Working Paper no. W01/22, The Institute for Fiscal Studies, London. Bond, Stephen and John van Reenen (2007): Microeconometric models of investment and Employment, Handbook of Econometrics, Volume 6A, pp 4417-4498, Holland: Elsevier. Dixit, A. and R. Pindyck (1994): Investment under Uncertainty, Princeton University Press, Princeton. Frankel, Jeffrey A. and Andrew K. Rose (1996): Currency crashes in emerging markets: An empirical treatment, Journal of International Economics, Vol. 41, no. 3-4, pp 351-366. Gezici, Armagan (2007): Investment under financial liberalization: channels of liquidity and uncertainty, PhD Thesis, University of Massachusetts Amherst. Harrison, A. and M. McMillan (2003): Does Direct Foreign Investment Affect Domestic Firms Credit Constraints?, Journal of International Economics, Vol. 61, No. 1, pp 73-100. Harrison, A., I. Love and M. McMillan (2004): Global Capital Flows and Financing Constraints, Journal of Development Economics, Vol. 75, No. 1, pp 269-301. Krugman, Paul (2000): Fire-Sale FDI, in Edwards, Sebastian (ed.): Capital Flows and the Emerging Economies: Theory, Evidence and Controversies, Mass: NBER. Moreno, Ramon, Gloria Pasadilla and Eli Remolona (1998): Asias Financial Crisis: Lessons and Policy Responses, Pacific Basin Working Paper Series, Center for Pacific Basin Monetary and Economic Studies, Economic Research Department, Federal Reserve Bank of San Fransisco. Ni, Shawn and Ronald A. Ratti (2009): Heterogenous Parameter Uncertainty and the Timing of Investment during Crisis, Economics E-journal, Discussion Paper, No. 2009-12, Kiel Institute for the World Economy, Germany. Serven, Luis (1998): Macroeconomic Uncertainty and Private Investment in LDCs: An Empirical Investigation, unpublished Working Paper, Washington D.C.: The World Bank.

12

APPENDIX TABLE A1. Data for the uncertainty variables Year WPI-SDA RiskE LOW HIGH LOW LOW LOW LOW LOW LOW HIGH LOW LOW LOW LOW LOW LOW LOW ERUSD-SDB 0.07 0.51 0.43 0.13 0.05 0.13 0.06 0.11 0.36 0.32 0.16 0.11 0.00 0.07 0.10 0.17 RiskF LOW HIGH HIGH LOW LOW LOW LOW LOW HIGH HIGH HIGH LOW LOW LOW LOW HIGH RiskPC HIGH HIGH LOW LOW HIGH LOW LOW LOW LOW LOW LOW LOW LOW LOW HIGH LOW

1993 0.80 1994 6.80 1995 1.33 1996 1.34 1997 1.23 1998 1.23 1999 1.54 2000 1.21 2001 3.25 2002 0.93 2003 0.91 2004 0.94 2005 0.38 2006 0.71 2007 0.74 2008 1.65 Explanatory notes A

Economic uncertainty is measured with inflation volatility. Based on the monthly wholesale price index published by the Central Bank (www.tcmb.gov.tr) is calculated the monthly rate of inflation. The standard deviation reported here refers to this standardised measure of volatility. Economic uncertainty is classified to be high when this measure is unusually high (e.g. above 3). Financial uncertainty is measured with the TL-USD exhcange rate. The Central Bank (www.tcmb.gov.tr) publishes daily exchange rates. From this Exchange rate is calculated the average annual volatility over the year. The Standard deviation reported here refers to this standardised measure of volatility. Financial uncertainty is classified to be high when this measure is unually high (e.g. above 0.16). Political uncertainty is measured using the Civil Rights Liberty index for Turkey published by Freedom House (www.freedomhouse.org). Political uncertainty is classified to be high when civil liberties decline (this happens only in 1993-1994). To this is added a high score for 1997 and 2007 where Turkey underwent significant crisis in its political system (28 February and Ergenekon).

13

TABLE A2. Descriptive Statistics and Pearson Correlation Coefficients Panel A Unbalanced panel, 1993-2007, 709 firms The table reports mean followed by SD below and correlation coefficients referring to row numbered variables 1. 1. I/K (rate) 2. Q (ratio q-1!) 3. ROA (ratio) 4. K (TL, 1993 prices) 5. Sales ( TL, 1993 prices) 6. Labour (no. of workers) 0.14 1.05 0.58 5.09 0.08 0.25 2*106 6*106 4*108 2*109 1,133 2,265 1 0.02 0.02 0.01 -0.007 0.005 1 1 -0.006 0.01 -0.13 1 -0.006 0.01 -0.13 1 0.21 0.61 1 0.22 1 2. 3. 4. 5. 6.

14

TABLE A2. Descriptive Statistics and Pearson Correlation Coefficients Panel B Unbalanced panel, 1999-2008, 491 firms The table reports mean followed by SD below and correlation coefficients referring to row numbered variables 1. 1. I/K (rate) 2. Q (ratio q-1!) 3. ROA (ratio) 4. K (TL, 1999 prices) 5. Sales (TL, 1999 prices) 6. IA (TL, 1999 prices) 0.05 0.29 0.51 2.59 0.08 0.13 2*107 5*108 9*107 2*109 2*105 6*106 1 0.09 0.09 0.04 0.03 0.03 1 1 0.01 0.03 0.01 1 0.01 0.03 0.01 1 0.89 0.76 1 0.73 1 2. 3. 4. 5. 6.

15