Exchange Rate Volatility and the Timing of Foreign Direct Investment: Market-seeking versus Export-substituting

Transcription

1 Exchange Rate Volatility and the Timing o Foreign Direct Investment: Market-seeking versus Export-substituting Chia-Ching Lin, Kun-Ming Chen, and Hsiu-Hua Rau * Department o International Trade National Chengchi University Abstract This paper examines the impact o exchange rate uncertainty on the timing o oreign direct investment (FDI) with heterogeneous investing motives. We irst extend Dixit- Pindyck s real options model to show that while an increase in exchange rate volatility tends to delay the FDI activity o a market-seeking irm, it might accelerate the FDI activity o an export-substituting irm i the irm s degree o risk aversion is high enough. The rationale behind this inding is that a market-seeking FDI might increase the exposure o the irm s proits to exchange rate risk, while an export-substituting FDI might reduce it. Firm-level data on the entry by Taiwanese irms into China over the period between 1987 and 00 are used to test the theory s validity. Empirical evidence rom a survival analysis based on the data is consistent with the theory. These results reveal that the relationship between exchange rate uncertainty and FDI is crucially dependent on the motives o the investing irms. Hence, it is essential to consider this actor in an empirical model so that the testing results are ree rom aggregation bias. Keywords: FDI, exchange rate, real options, survival analysis JEL Classiication: F1, F31, G13 * Corresponding author: Kun-Ming Chen, Tel: kmchen@nccu.edu.tw. Chen and Rau are Associate Proessors and Lin is a graduate student at the Department o International Trade, National Chengchi University, Taipei, Taiwan.

2 1. Introduction Ever since the breakdown o the Bretton Woods system in the 1970s, the exchange rates o many countries have been luctuating considerably over time. A large body o recent research deals with the implications o exchange rate uncertainty or the real economy. Regarding the eects o exchange rate uncertainty on oreign direct investment (FDI, hereater), while many theoretical and empirical studies indicate that exchange rate volatility has had a signiicant eect on FDI movements, the impact o exchange rates is ound to be heterogeneous across countries and types o investment, and vary over time. 1 Previous theoretical studies demonstrate that exchange rate volatility aects FDI activity through two main channels: irms attitude towards risk and the option value o investment lexibility. It has been suggested that, or a risk-averse irm, higher volatility lowers the certainty equivalent value o the investing irm. Hence, FDI decreases as exchange rate volatility increases. By contrast, Itagaki (1981), Cushman (1985), and Goldberg and Kolstad (1995) illustrate the importance o considering the post-fdi changes in the exposure o a irm s proits to exchange rate risk. I the investing irm can choose to serve oreign markets via exports or FDI, then an increase in exchange rate volatility might lead the irm to substitute FDI or exports, since FDI activity reduces the exposure o its proits to exchange rate risk. The studies mentioned above are based on the traditional investment theory which assumes that an investment decision is to be taken now or never. They ignore the option o delaying an investment. Beginning in the 1980s a real options theory has been developed to analyze investment behavior. Under the assumptions o uncertainty and irreversible 1 For instance, Amuedo-Dorantes and Pozo (001), Bell and Campa (1997), Campa (1993, 1994), Darby et al. (1999), Crowley and Lee (00), and Kiyota and Urata (004) and Chen et al. (005) ind that exchange rate uncertainty deters FDI activity. However, Cushman (1985), Goldberg and Kolstad (1995), and Pain and Van Welsum (003) illustrate that exchange rate stimulates FDI lows. See, or instance, Wihlborg (1978). 1

3 investment, the real options theory emphasizes the option value o the lexibility that a irm has in possibly delaying an investment decision in order to obtain more inormation about the uture. Dixit (1989a,b) indicates that the waiting value increases as the uncertainty rises even or a risk-neutral irm. Hence, an increase in exchange rate uncertainty will deer the FDI activity o the irm. Using Dixit-Pindyck s (1994) model, however, Darby et al. (1999) illustrate that, or a risk-averse irm, the impact o exchange rate uncertainty on the timing o FDI is ambiguous. A limitation o Dixit-Pindyck (1994) and Darby et al. (1999) is their treatment o irms risk aversion. The risk aversion is incorporated into their model through a risk premium added to the private discount rate. This approach ignores an important eature in the traditional theory that allows the exposure o the investing irm s proits to exchange rate risk to vary with dierent types o FDI. 3 To ill the gap in the literature, the purpose o this paper is to reexamine the relationship between exchange rate uncertainty and FDI both theoretically and empirically. We irst develop an integrated ramework o FDI under uncertainty in which a irm s attitude towards risk and the option value o investment lexibility are incorporated simultaneously. In this regard, Dixit-Pindyck s (1994) real options model is extended to consider possible changes in the post-fdi exposure to exchange rate risk. It is shown that the relationship between exchange rate uncertainty and FDI varies with the extent o the exposure to exchange rate risk which is determined by investing motives. This paper inds that exchange rate volatility tends to delay the FDI activity o a market-seeking irm, but it may accelerate the FDI activity o an export-substituting irm. We use irm-level data on Taiwan s outbound FDI in China over the period to test the validity o our theoretical results. 3 Although Erdal (001) and Chen et al. (005) consider dierent motives o irms, the risk neutrality assumption in their model makes their results same as Dixit s analysis. The exposure problem also has not been discussed in their ramework.

4 Empirical evidence rom a survival analysis based on the data is consistent with the theory. The remainder o the paper proceeds as ollows. In the ollowing section, Dixit- Pindyck s (1994) real options model is extended and the eects o exchange rate volatility on the FDI activity o market-seeking irms versus export-substituting irms are illustrated. Section 3 discusses our empirical methodology and model, ollowed in the subsequent section by a presentation o the data and empirical results. Brie concluding remarks are given in the inal section.. A simple model o FDI under exchange rate uncertainty Dixit-Pindyck s (1994) real options model is extended here to reexamine the relationship between exchange rate uncertainty and FDI. To illustrate the importance o the diversity o motives in investigating the determinants o the timing o FDI, we ocus on two extreme cases according to motives o investing irms, namely, export-substituting FDI and market-seeking FDI. Export-substituting FDI reers to the situation in which an exporting irm, originally producing at its home country and serving a oreign market via exports, relocates its whole production abroad to serve the oreign market. 4 The motive or export-substituting FDI is to reduce the irm s production costs. By contrast, market-seeking FDI reers to the situation in which a domestic irm, originally not serving a oreign market via exports, chooses to set up a oreign subsidiary to produce and sell in a given oreign market. Thus, the motive o market-seeking FDI is to create a new market or its product. Suppose that the irm is a price taker and it produces a unit low o output at ixed marginal cost per period. For simplicity, we assume that the variable costs comprise labor cost only and the input-output coeicients are ixed to be one. Thereore, the variable costs can be treated as the wage rate. Finally, it is assumed the exchange rate R, expressed in units o home 4 The conclusion remains the same even i we allow the irm to substitutes FDI or exports partially. 3

5 currency per oreign currency, ollows an exogenously geometric Brownian motion dr R = µ dt + σ dz. (1) Here, µ is the growth rate o the exchange rate; σ is the volatility o the exchange rate; t is the time path and z is a Wiener process. 5 The objective o the irm is assumed to obtain maximum expected utility in terms o its home country s currency. In order to introduce the concept o risk aversion and associate it with the motives o the irm, the ollowing mean-variance expected utility unction 6 is adopted: 1 EU ( π ) = E[ π] a Var( π ) () p where EU (.) is the expected utility; π shows proits o the irm; a p is Arrow-Pratt s absolute risk aversion coeicient; Var(.) is the variance. Note that µ Var( ) (.) 0 t σ t R e ( e 1) π =, where R 0 is initial value o exchange rate and (.)>0 is a unction o π. It is obvious that a rise in σ or µ will increase the variance o the proits. Export-substituting FDI In the case o export-substituting FDI, it is assumed that there are two possible entry modes or the domestic irm to serve a oreign country: Export versus FDI. Suppose that prior to implementing FDI, a irm produces goods at its home country and exports products to a oreign country. Hence, its proit lows in terms o home country s currency per period are 0 π = PR Wd (3a) where π is the proit unction; the superscript 0 represents the pre-fdi state; P is oreign market price in terms o oreign currency and W d is domestic wage rate. Ater the irm invests 5 The subscript t o R and π is suppressed in this section or simplicity. 6 This unctional orm is used by Kawai (1984) and Qin (000). Moreover, Cushman (1985) and Goldberg and Kolstad (1995) also use a similar setting to analyze the impact o exchange rate uncertainty on FDI. 4

6 to produce abroad and serve the oreign country rom its oreign subsidiary 7, its proit lows become 1 π = PR WR (3b) where superscript 1 represents the post-fdi state; W is oreign wage rate in terms o oreign currency. From Equations (3a) and (3b), it is obvious that i the irm substitutes FDI or exports, the exposure o its proits to exchange rate risk will be reduced due to the act that ( PR WR) < PR. From Equations (1)-(3), the change in expected utility, EU ( π ), rom substituting FDI or exports can be derived as ollows: ( ) ( ) ( ) [ ] ( ) ( ) 1 0 µ t σ t π π π d 1 EU = EU EU = W W E R a P W P e e (4) 1 where a = a R 0. From Equation (4), i the irm invests to produce at the oreign country p and stays in the market orever, the change in expected present utility, ξ E, becomes WR ρ t Wd ( ) ( ) ( ) ξe R = EU π e dt a P W P = + γ 0 ρ µ ρ (5) where ρ is the irm s discount rate; γ σ ρ µ σ ρ µ = [( )( )] ; subscript E represents an export-substituting irm. For the purpose o convergence, we assume ρ µ σ > +. It is obvious rom Equation (5) that a depreciation o the home country s currency (i.e., an increase in R ) causes a reduction inξ ( R), thus deterring its irms FDI activity. E The decision problem o the irm is to choose an optimal time to enter the oreign market. At time t, the irm can either produce in the host country ater investing a lump sum k and gets the extra expected present utility as shown in Equation (5), or stays in the original state and keeps the right to invest in the next period. Hence, in each period the irm aces a binary 7 We assume that the total output o the oreign subsidiary is sold in the oreign country. The reverse-import case is excluded in the model. 5

7 decision problem as ollows: 1 V( R) = max ξ E ( R) k, E V( R ) R 1 + tρ (6) where V is the optimal expected net present value; t is the time interval k is the sunk costs expressed in the home country s currency; R is the exchange rate in period t+1. The ormer term on the right-hand side, ξ ( R ) k, is the net entry value, and the latter term, 1 (1 + tρ) E[ V( R ) R], is the value o the option to wait. E t Since the utility unction in this case is a decreasing unction o R, there is a cuto point, R L, at which i R<R L, then net entry value ξe ( R) k is greater than the value o the option to wait. Thus, the irm s optimal decision is to carry out FDI. Using value-matching and smooth-pasting conditions, we have ρ µ α Wd ( ) RL = a P P W γ k W 1 α + ρ + (7) where α σ µ σ µ σ σ ρ = [( 0.5 ) + ( 0.5 ) + ] > 0. 8 The higher the value o R L is, the higher the probability will be that R is smaller than R L. Hence, the irm has higher incentive to invest earlier. To ensure that there is a possibility or a risk-neutral irm to undertake FDI, we assume that W ρ k > 0. d In the ollowing, beore we discuss the general case presented in Propositions 4 and 5, we irst derive the results o two special cases which correspond to the speciications in previous studies. The irst special case is an investment decision o a risk-averse irm that has to be taken now or never; that is, there is no option to delay the investment. The other special case is an investment decision o a risk-neutral irm with an option to delay the investment. 8 See Dixit (1989b), p.66. 6

8 Proposition 1 In the case o export-substituting FDI, an increase in exchange rate volatility will stimulate FDI activity o a irm i the irm cannot delay its investment. Proo: From Equation (5), it can be shown that ξe ( ) σ = a P P W 0 σ >, ( ρ µ σ ) which implies that the irm has a higher incentive to substitute FDI or exports i the exchange rate uncertainty rises. The economic rationale behind Proposition 1 is straightorward. I the irm cannot delay its investment, then the risk attitude is the only channel through which exchange rate uncertainty aects FDI. Substituting FDI or exports reduces the irm s exposure to exchange rate risk, and this gain rom risk reduction is larger i the exchange rate is more volatile. Consequently, an increase in exchange rate volatility stimulates the irm s FDI activity. This result is similar to that ound in Itagaki (1981). Proposition A risk-neutral export-substituting irm will delay its FDI activity when the exchange rate volatility rises; that is, R σ = 0< 0. L a p Proo: From Equation (7), we have RL RL α = < 0 σ α + α σ ap = 0 ( 1 ) where α α µ + σ = [1 + ] < 0. σ σ ρσ + ( µ 0.5 σ ) The economic intuition o Proposition is as ollows. I the irm is risk-neutral, then the option value o investment lexibility is the only channel through which exchange rate uncertainty aects FDI. An investment is like a call option whose value rises i the underlying 7

9 uncertainty increases. Hence, acing an irreversible investment and uncertain uture, a potential entrant has more incentive to delay its investment so as to get extra inormation. Proposition 3 A risk-neutral export-substituting irm will delay its FDI activity under a rising exchange rate trend; that is, R µ = 0 < 0. L a p Proo: From Equation (7), we have RL RL ψ = µ + α ρ µ ρσ + µ σ ) ap = 0 ( 1 )( ) ( 0.5 < 0. where ψ ρ µ α ρσ µ σ = (1 + ) + ( 0.5 ) < 0. 9 The reason or µ to be negatively related to the FDI activity o a risk-neutral irm is due to the act that µ represents the expected uture level o exchange rate. Moreover, a greater level o µ implies that the probability o uture level o exchange rate being less than R L is smaller. Hence, the risk-neutral irm will delay its investment as µ rises. Lemma 1 RL σ ap > 0 and RL µ ap > 0. Proo: See Appendix 1. Proposition 4 In the case o export-substituting FDI, the eect o exchange rate volatility on the timing o FDI is ambiguous. However, there exists a threshold in the degree o risk aversion a% such that this eect is positive (negative) i the irm s risk-aversion coeicient a is greater (smaller) than a%. p 9 See Chen et al. (005, p.15) or the proo o ψ < 0. 8

10 Proo: From Equation (7), we have RL =Γ 1+ Γ σ ap (8a) where and ρ µ 1 Wd Γ 1 = α k W (1 + α) ρ σ R ρ 0 µ α P ( P W ) γ α σ Γ = W 1 α + α(1 α) σ ( ρ µ σ. + + ) Given RL σ a = 0=Γ 1 < 0 (Proposition ) and RL σ ap =Γ > 0 (Lemma 1), since p R σ is a linear unction o a, there must exist a critical value, a%, at which R σ > 0 L L i a > %a, and R σ < 0 i a < %a. p L p The economic intuition o Proposition 4 is as ollows. As shown in Propositions 1 and, exchange rate volatility σ aects the FDI through two channels: the risk attitude o a irm and the option value o investment lexibility. These two channels have opposite eects on the FDI activity o an export-substituting irm. Thereore, the eect o exchange rate volatility on FDI is ambiguous. However, given the negative eect on FDI activity rom the option value o investment lexibility, i the positive eect resulting rom the risk aversion as well as the change in the exposure to exchange risk resulting rom the irm s FDI becomes large (small) enough, the net eect will be positive (negative). Proposition 5 In the case o export-substituting FDI, the eect o the exchange rate trend on the timing o FDI is ambiguous. However, there exists a threshold in the degree o risk aversion â such that this eect is positive (negative) i the irm s risk-aversion coeicient a p is greater (smaller) than â. 9

11 Proo: From Equation (7), we have RL =ϒ 1+ apϒ µ (8b) where and ψα Wd ϒ 1 = k (1 α) W ρσ ( µ 0.5 σ ) ρ + + and R 0 ( ρ µα ) P ( P W ) ψγ γ ϒ = + W (1 ) + α (1 + α)( ρ µ ) ρσ + ( µ 0.5 σ ) µ γ σ (ρ 4 µ σ ) = > 0. µ ( ρ µ σ ) ( ρ µ ) Given RL µ a = 0= ϒ 1 < 0 (Proposition 3) and RL µ ap =ϒ > 0 (Lemma 1), since p R L µ is a linear unction o a, there must exist a critical value, â, at which R L µ > 0 i a > aˆ, and R µ < 0 i a < aˆ. p L p Regarding the eect o exchange rate trend µ, on the one hand, because µ represents the expected uture level o exchange rate, an increase in µ decreases the probability o uture level o exchange rate being smaller than R L. Hence, the irm will delay its investment, as shown in Proposition 3. On the other hand, an increase in µ will raise the variance o a irm s proits as mentioned above, and thus stimulate the FDI activity o a risk-averse export-substituting irm. Thereore, the eect o exchange rate trend on FDI is also ambiguous. However, the latter eect is larger or a higher risk-averse irm, as shown in Lemma 1. Thus, when the degree o risk aversion exceeds a critical level, the latter positive eect will dominate the ormer negative eect, meaning an increasing trend in exchange rate will stimulate FDI activity o an export-substituting irm, and vice versa. 10

12 Market-seeking FDI Since we assume a market-seeking irm has not served the oreign market via exports prior to undertaking FDI, the proits rom exports are zero in state 0. The change in net present utility, ξ M, rom FDI can be shown as ollows 10 : ξ M ( ) ( ) P W R R = a P ρ µ ( W ) γ (9) where subscript M represents the market-seeking irms. It is obvious rom Equation (9) that a depreciation o the home country s currency (i.e., an increase in R ) will raise the value o ξ M ( R), thus stimulating its irms FDI activity. The binary decision problem or the irm in each period is 1 V( R) = max ξ M ( R) k, E V( R ) R 1 + tρ (10) There is an entry threshold rate R H at which a potential entrant enters i R>R H. In other words, the lower the value o R H is, the higher the incentive will be or the irm to enter the market. Using value-matching and smooth-pasting conditions, we have ρ µ β R ( ) H = a P W γ + k P W β 1 (11) where β σ µ σ µ σ σ ρ = [ ( 0.5 ) + ( 0.5 ) + ] > Proposition 6 In the case o market-seeking FDI, an increase in exchange rate volatility will delay the FDI activity o the irm; that is, R σ > 0. H Proo: From Equation (11), we have 10 The derivation o this equation is similar to the case o the export-substituting FDI. 11 See Dixit (1989b), p

13 RH RH β ρ µ βa σ = + P W σ β β 1 σ P W β 1 ( ρ µ σ ) ( ) ( ) > 0, (1a) where β β µ + σ = [ 1 + ] < 0, given the assumption σ σ ρσ + ( µ 0.5 σ ) ρ µ σ > +. The economic intuition o Proposition 6 is as ollows. In this case, FDI activity will make the irm s exposure to exchange rate risk increase. Thus, an increase in exchange rate volatility will reduce the expected utility gain rom this activity or a risk-aversion irm. At the same time, an increase in exchange rate volatility will increase the option value o delaying the investment so as to deter the FDI activity urther. Proposition 7 A risk-neutral market-seeking irm will accelerate its FDI activity when the exchange rate trend rises, that is, R µ = 0 < 0. H a p Proo: From Equation (11), it can be shown that RH RH φ = < 0 µ ap = 0 ( β 1)( ρ µ ) ρσ + ( µ 0.5σ ) where φ ρ µ β ρσ µ σ = ( 1) + ( 0.5 ) < 0. 1 Similar to Proposition 3, the economic intuition o Proposition 7 is straightorward. An increase in µ raises the probability o uture exchange rate level to be larger than R H, and thus stimulates the risk-neutral irm to invest earlier. Lemma RH µ ap > 0. 1 See Chen et al. (005, p.14) or the proo o φ < 0. 1

14 Proo: See Appendix 1. Proposition 8 In the case o market-seeking FDI, the eect o exchange rate trend on the timing o FDI is ambiguous. However, there exists a threshold in the degree o risk aversion a such that this eect is negative (positive) i the irm s risk-aversion coeicient a p is greater (smaller) than a. Proo: From Equation (11), we have RH =Λ 1+Λa µ p, (1b) where Λ = 1 φβ k β P W ρσ + µ σ ( 1) ( ) ( 0.5 ) and Here and R Λ = + ρ µ β 0 ( P W ) φγ γ P 1 W β ( β 1)( ρ µ ) ρσ + ( µ 0.5 σ ) µ R H µ is a linear unction o a. In addition, RH =Λ 1 < 0 (Proposition 7) µ a p = 0 RH µ ap =Λ > 0 (Lemma ), which must exist a trigger value a at which R H µ > ( < )0 i ap > ( < ) a. The reasoning regarding the eect o exchange rate trend in the market-seeking case is similar to what we have ound in the export-substituting case. An increase in µ aects FDI through two channels: the probability o R being greater R H and the variance o the irm s proits. The eect o an increase µ on FDI activity rom the irst channel is positive (Proposition 7), and the eect rom the second channel is negative, its total eect is ambiguous. However, when the degree o risk aversion exceeds a trigger level, the second 13

15 eect will dominate the irst eect, thus an increasing trend in exchange rate will deer FDI activity o a market-seeking irm, and vice versa. Proposition 9 Sunk costs are negatively related to the FDI activity o both export-substituting irms and market-seeking irms; that is, R k < 0, and R k > 0. Proo: The proos are straightorward, and thus are omitted. L H The economic intuition o Proposition 9 is clear that the higher the entry costs are, the higher the revenues or the lower the variable costs will be that are requested to compensate or the opportunity loss. Proposition 10 The FDI activity o export-substituting irms is negatively related to oreign wage rate and positively related to domestic wage rate; that is, R W < 0, R W > 0. Proo: From Equation (7), it can be shown that and RL ρ µ α 1 = > 0, W W 1+ αρ d. + RL ρ µ α Wd = 0 + aw k < W W 1 α ρ The second inequality holds due to the assumption o W ρ k > 0, mentioned in Equation (7). d L L d Proposition 11 The oreign wage rate is negatively related to the FDI activity o a risk-neutral market-seeking irm. However, the relationship between oreign wage rate and the FDI activity or a risk-averse market-seeking irm is ambiguous. Proo: From Equation (11), we have 14

16 RH ρ µ β = k a( P ) W W ( P W ) β 1 γ. It is obvious that the sign o RH W is ambiguous, depending on the relative magnitude o risk aversion, a p, and the sunk costs, k. However, i ap =0, then R W > 0. H An increase in domestic wage will make the variable production costs in the oreign country relatively cheap, and thus cause export-substituting FDI to increase. Thereore, the relation between domestic wage and export-substituting FDI is positive. As or the reason why the eect o oreign wage on the FDI activity o export-substituting irms is negative while its eect on the FDI activity o market-seeking is ambiguous is due to the ollowing act: on the one hand, the higher the oreign wage rate is, the higher the variable costs will be that are involved in oreign production which reduces the proitability o both types o FDI. On the other hand, an increase in the oreign wage rate will raise the exposure o the proits o an export-substituting FDI, but reduce the exposure o the proits o a market-seeking FDI. Thereore, the total eect o oreign wage rate on the FDI activity o export-substituting irms is negative, while its eect on the FDI activity o market-seeking is ambiguous. Nevertheless, as obvious rom the proo o Proposition 11, i the sunk costs are high, or the degree o risk aversion is low enough, then the total eect o oreign wage rate on the FDI activity o market-seeking is also negative. To sum up, we develop an integrated model o FDI under uncertainty to reexamine the relationship between exchange rate uncertainty and FDI activity with heterogeneous investing motives. We ind that while exchange rate volatility tends to deer the FDI activity o a market-seeking irm, it may accelerate the FDI activity o an export-substituting irm i the degree o risk aversion is high enough. We have shown that previous studies with either a real options model (such as Dixit (1989a, b), or a traditional risk-aversion model (such as Itagaki (1981), Goldberg and Kolstad (1995) and Cushman (1985)), are equivalent to the special 15

17 cases o our model. Our results demonstrate the importance in considering the diversity o investing motives when examining the relationship between exchange rate uncertainty and FDI activity. Table 1 summarizes the expected signs o the determinants o market-seeking FDI versus export-substituting FDI. 3. Empirical methodology and model 3.1 Empirical methodology This paper ocuses on the analysis o how exchange rate volatility aects the timing o oreign entry. One widely applied method to examine the issue about timing is to conduct event history analysis. Event history analysis investigates what may happen over a time span beore a certain event occurs. In our case, the event is a irm s entry into a oreign market. The waiting time or a irm to enter a oreign market can be treated as the survival time o the irm, and the timing o entry can be treated as the timing o event occurrence. To apply this method, one needs to speciy a survival unction to describe the probability o a irm s survival until a certain time has elapsed. The probability o a irm s entry at a certain time period can be expressed by a hazard unction. When we denote the probability density unction o event occurrence as () t, the hazard unction ht () can be written as ht () = () t St (), where St () is the survival unction which can be speciied as St () = Pr( T t), where T is the duration o survival o a irm and t is a certain time point. This paper adopts Cox s proportional hazard model (Cox (197, 1975)). 13 One o the advantages about Cox s model is that it imposes the condition o hazard proportionality and thus makes the analysis o covariates possible without speciying a hazard unction itsel. The 13 Cox model was originally developed in the ield o biology and medical science. It has been applied in economics and other social sciences since the mid 1990s, such as the survival time o corporate irms (e.g. Kimura and Fujii (003) and Van Kranenburg et al. (00)), the entry time o irms (e.g. Kogut and Chang (1996), Ursacki and Vertinsky (199), and Leung et al. (003)), and problems in management (e.g. Fuentelsaz et al. (00) and Tan et al. (00)) or political science (e.g. Box Steensmeier (1996) and Oneal and Russett (1997)). 16

18 model treats each sample s hazard rate hi ( t) as a unction o a number o covariates and conceptually deines the baseline hazard h () t 0 that is not inluenced by any covariate. Based on the hazard proportionality assumption, the model treats the proportion o hi ( t) and h () t 0 as constant. Hence, the proportion is interpreted as a unction o covariates. We deine the hazard rate as the rate at which a irm invests in a oreign country by t given that the irm has stayed in the home country until t. Thus, the hazard unction, ht ( x ), can be expressed as 15 ht ( x i1,x () t) = h0() texp( β x1 + α x () t i i i ) (13) 14 i where h () 0 t is the baseline hazard unction; β and α are p 1 and q 1 vectors, respectively; is a vector o time-independent covariates; x ( t) is a vector o time x1 i i varying covariates; subscript i represents i th irm; subscript t represents time. Suppose that we have a dataset with n observations and K distinct entry times. I we sort the sample by the order o entry times, then the partial likelihood unction,, becomes L p L p = δ i n β x1 i + α x i( t) e β x1j α (14) + x j ( t ) i= 1 e j Ω( ti ) where Ω( t i ) represents the number o irms that are at risk o experiencing an entry at time ti, that is, the risk set ; δ i is an indicator, its value is 0 i the sample is right-censored, and 1 i the sample is uncensored. 16 The positive (negative) estimators ˆβ and ˆα represent the variables have positive (negative) impacts on the occurrence o the event. To solve Equation 14 See Box-Steensmeier and Jones, (004), Chapters and For more details, see Lawless (003) or Box-Steensmeier and Jones (004). 16 I irms do not invest in the sample period but may invest in the uture, then the sample is reerred to as a right-censored sample. 17

19 (14), there are three methods to compute the ties 17 : Breslow method, Eron method and Exact discrete method 18. It turns out that our results are not sensitive to which method is used. 3. Empirical model Based on the theoretical ramework o this paper and Equation (13), the ollowing empirical model is established: [ ] log ht ( ) h( t) = ασ + α EX * σ + α EX * σ 0 1 t M t 3 E t + α R + α EX *R + α EX *R 4 t 1 5 M t 1 6 E t 1 + αµ + α EX * µ + α EX * µ 7 t 8 M t 9 E t + α10wage + t 1 α11 PFi, ti + β MKT + β FUND 1 + β R & D + β SIZE + β SIZE * SIZE 3 i 4 i 5 i i + β KL + β HT 6 i 7 i i (15) Here, subscript i represents i th irm and subscripts M and E represent the market-seeking irm and export-substituting irm, respectively. Since Taiwanese irms were not permitted to invest in China until 1987, the dependent variable is deined as the duration rom 1987 to the year when the irm invested there. As or independent variables, in addition to the variables suggested in Section, some are added as explanatory variables in order to control or some important actors that are not considered in our theoretical ramework. The deinition o these variables and their expected signs are discussed as ollows (see also Table ): σ t : exchange rate volatility. According our model in Section, while exchange rate uncertainty tends to deter the FDI activity o market-seeking irms, its impact on export-substituting irms is ambiguous. To test the validity o our theory, we deine two dummy variables: 1. EX M, whose value is 1 or market-seeking irms, and 0 otherwise;. EX E, whose value is 1 or export-substituting irms, and 0 otherwise. Thereore, the expected 17 Ties occur when two or more irms enter a market at the same observed time. 18 See Box-Steensmeier and Jones (004) p.54 18

20 sign o ( α 1 + α ) is negative, and that o ( α 1 + α 3 ) is positive (negative) or those export-substituting irms with high (low) risk-aversion. R t-1 one-period lagged real exchange rate o NTD versus RMB, in which nominal exchange rates are delated with prices o the respective countries to control or the possible movements in prices ollowing the change in nominal exchange rates. Since it is time-consuming to make an FDI decision, the inal decision might be more related to the previous exchange rate level, and thus the one-period lagged values are used. According to our model, an appreciation o the host country s currency increases market-seeking irms proits in terms o the home currency and decreases those o export-substituting irms. In the empirical equation we also use the dummy variables EX M and EX E to test the validity o our theoretical results. The expected sign o ( α 4 + α 5 ) is positive, and that o ( α 4 + α 6 ) is negative. µ t : trend o exchange rate. According to our theoretical ramework, or irms with very low risk-aversion, an increase in µ t accelerate the FDI activity o market-seeking irms and delay the FDI activity o export-substituting irms. By contrast, or irms with very high risk-aversion, an increase in µ t delays the FDI activity o market-seeking irms and accelerates the FDI activity o export-substituting irms. Thereore, i the risk- aversion o the irms is very low, then the expected sign o ( α7 + α8) is positive and that o ( α7 + α9) is negative. I the risk-aversion o the irms is very high, then the expected sign o ( α7 + α8) is positive and that o ( α7 + α9) is negative. W W : wage rate o the oreign country relative to that o the home country. The ratio d o China s one-period lagged real wage rate over Taiwan s one-period lagged real wage rate (WAGE t-1 ) is used. According to our theory, the expected sign o the coeicient or export-substituting irms is negative, and that or market-seeking irms is ambiguous. 19

21 MKT i : marketing intensity, a proxy variable o the sunk costs. According our theory, the expected sign o its coeicient is negative. As or the control variables, based on the previous studies, 19 the ollowing variables are used: proits (PF), source o unds (FUND), R&D intensity (R&D), irm s size (SIZE), capital-labor ratio (KL), and high-tech industry dummy (HT). According to the liquidity hypothesis, since the cost o internal unds is viewed by investors to be lower than the costs o external unds, 0 there is a positive relation between a irm s internal cash lows and its investment abroad. The proit rate (PF i,t ) is used as a proxy o a irm s internal capital, the expected sign o its coeicient is positive. In addition, i the parent company o an investing irm can provide necessary unds, it suggests that the irm is more unlikely to ace inancial constrain, and thus it will be more likely to enter oreign markets earlier. We create a dummy variable FUND i, whose value is 1 i the parent company provides the necessary unds and 0 otherwise. We expect the sign o its coeicient to be positive. The internalization hypothesis indicates that due to high transaction costs o intangible assets, an investing irm with superior knowledge and management expertise will choose to set up a subsidiary rather than simply licensing a oreign irm to produce the product. R&D intensity variable (R&D i ) is used as a measure o the investing irm s intangible asset. The expected sign o this variable is positive. 1 In addition, Horst (197) argues that a irm s success at home will be highly correlated with its success abroad, since both are the result o the same technological and marketing capabilities. Hence, lager irms are more likely to invest abroad than smaller irms. The sales o a irm (SIZE i ) are used to measure its size and its expected sign is positive. However, as pointed out by Tan et al. (00), very large Taiwanese irms suer more rom institutional pressures rom Taiwan s government to not 19 See, or instance, Agarwal (1980), and Blonigen (005) or literature surveys. 0 This may be caused by imperections in the inancial and capital markets. 1 See also Blonigen (005). 0

22 invest into China due to hostility across the Taiwan Strait. Thus, the eect o the irm s size should have an inverse U-shape. That is, the expected sign o the coeicient o SIZE i is positive but that o SIZE i *SIZE i is negative. It has been suggested that one o the important driving orces behind FDI is to seek a production location with low labor cost (Kojima (1973)). Since the wage rate in China is signiicantly lower than that in Taiwan, a labor-intensive irm will beneit more rom investing in China. As a result, we expect that a irm s capital-labor ratio (KL i ) will be negatively related to its FDI activity. Finally, according to the OLI paradigm proposed by Dunning (1977), one o the three necessary conditions or a irm to undertake oreign direct investment is ownership advantage. Since Taiwan, relative to China, has a comparative advantage in high-tech industries (particularly the IT industries), these industries are more likely to expand their markets through FDI into China. However, the policy o to take root in Taiwan restricts high-tech industries investment timing, types and amount in China. We deine a high-tech dummy variable HT i which takes a value 1 or the high technology industries, and 0 otherwise. The expected sign o its coeicient is ambiguous. 4. The data and empirical results 4.1 The data The data on the dependent variable used in this paper are compiled rom the Survey on Taiwanese Firms in Mainland China, published by Taiwan s Investment Commission, Ministry o Economic Aairs (MOEAIC) in 003 and 004. It investigated all irms which invested in China or more than one year. This paper chooses 198 listed companies on Taiwan Stock Market rom the sample o the survey. Taiwan Economic Journal (TEJ) database indicates that among 1,145 available listed companies on Taiwan s Stock Market, 67 companies invested in China beore 00. Thus, our sample irms account or 9.5% o all 1

23 Taiwanese irms investing into China. To avoid the problem o sample selection bias, we use a uniorm distribution to randomly choose 139 companies rom those listed irms that had not invested in China beore 00. Because these companies might have invested in China ater 00, they are treated as right-censored samples. Thereore, our inal sample consists o 337 irms. Taiwan government prohibited domestic irms rom having any trade or investment relationship with China beore These barriers in trade or investment were removed or lowered starting in We thereore analyze the timing o Taiwanese irms entry into China during the period rom 1987 to 00. In other words, Taiwanese irms enter the risk set o entry rom 1987, but there are 36 irms that were set up ater 1987, and thus the risk set o these irms begins rom the years o their establishment. The entry years are obtained rom the government s oicial survey data. Figure 1 shows the distribution o entry years. The number o entries has increased considerably rom 1993, reaching a peak in 000. As mentioned above, in order to test the validity o theoretical prediction, two dummy variables EX M and EX E are created. We separate the sample irms into three groups, according to their pre-fdi export ratios: market-seeking irms (irms with zero exports), export-substituting irms (irms with export ratios more than 80%), and other irms. The deinition o these two variables are accordingly: (1) market-seeking irm dummy, EX M, taking a value 1 or a irm with zero exports i the sales o the irm s subsidiary account or more than 80% o its total sale in China, and 0 otherwise; () export-substituting irm dummy, EX E, taking a value 1 or a irm with a export ratio greater than 0.6 i the sales o its subsidiary account or more than 80% o the subsidiary s total sale in China, and 0 otherwise. According to these criteria, we have 3 market-seeking irms and export-substituting irms in our sample. To test the robustness o our empirical results, we have relaxed our criteria about investing irms by including

24 Several measures o trend and volatility o the real exchange rate have been proposed in the literature. Following Tsay (00, p.9), we irst use a modiied average and a modiied standard deviation o the monthly change in the logarithm o the real exchange rate to stand or the trend and volatility o the real exchange rates, which are designed to approximate a continuous-time geometric Brownian motion process. We then use a GARCH process to estimate the conditional mean and variance o the real exchange rate as the other measures o its trend and volatility, since some studies such as Pozo (199) note that exchange rates oten exhibit persistent behavior. 3 See Appendix or details about the calculation o these measures. Table 3 summarizes the distribution o the sample irms by industry. The electronics and electric industries account or signiicant shares o all sample irms as well as investing irms. Both shares are around 40%. Furthermore, most o export-substituting irms belong to electronic and electric appliances industries, but by contrast, most o market-seeking irms belong to services and ood & beverage industries. It is worth noting that the export ratios o the sample irms scatter widely with a standard deviation o 30.8%, which allows us to separate our sample irms into three groups so as to test the validity o our theory. Summary statistics o these variables are summarized in Table Empirical results Table 5 summarizes the estimation results o our empirical model. 4 The regression equations reported in columns 1 and 3 are our benchmark case, in which the dummy variables EX M and EX E that control or investing motives are not considered. In columns and 4 we introduce these two dummy variables in order to test the validity o our theory. those irms with sales o its subsidiary accounting or more than 50% o the subsidiary s total sales in China. We ind that the empirical results are basically the same. 3 The measure o Tsay (00) belongs to unconditional variance, and the measure o GARCH belongs to conditional variance. 4 The estimation in Table 5 uses the method o Eron. 3

25 In column 1, both o the coeicients o σ t and µ t are positive, but not statistically signiicant, and the coeicient o R t-1 is signiicantly negative. However, the results in column reveal that there is considerable heterogeneity in the eects o the determinants o FDI among dierent types o FDI. The eect o real exchange rate volatility on market-seeking FDI, as shown in the joint test o ( α1 + α ), is signiicantly negative, while its eect on export-substituting FDI, as shown in the joint test o ( α1+ α3), is signiicantly positive. As or the eect o real exchange rate, whereas its eect on market-seeking FDI, as shown in the joint test o ( α4 + α5), is signiicantly positive, its eect on export-substituting FDI, as shown in the joint test o α + α ), is signiicantly negative. It is also worth noting that, in contrast to the insigniicant ( 4 6 result in column 1, the eects o real exchange rate trend on market-seeking FDI and export-substituting FDI, as shown in the joint tests o ( α7 + α8) and ( α7 + α9), are both signiicantly positive. These results are consistent with the prediction o our theory. It also demonstrates that testing results without considering the heterogeneity in the investing motives might suer rom aggregation bias. As or the control variables, the coeicients o irm s size (SIZE i ), source o unds (FUND i ), proit (PF i,t ), and R&D intensive (R&D i ) are signiicantly positive. This indicates that Taiwanese irms that have a larger size, are unding rom their parent companies, have higher proit rates, and have a higher R&D intensity tend to have higher incentive to invest into China earlier. Furthermore, the coeicient o SIZE i * SIZE i is signiicantly negative, which suggests that the entry o very large Taiwanese irms into China might be deterred due to Taiwan government s policy. In addition, the coeicient o KL i is signiicantly negative and the coeicient o HT i is negative but not statistically signiicant, indicating that investing irms in labor-intensive or traditional industries are more likely to invest into China. In general, these results are consistent with previous studies. Finally, the results in columns 3 4

26 and 4 show that the empirical results in columns 1 and are not qualitatively sensitive to dierent measures o the trend and volatility o real exchange rates. To evaluate the relative importance o these covariates in the entry decision o investing irms, a useul ormula is given by β( xi = X1 ) β( xi = X) e e % ht () = 100, 5 β ( xi = X) e where x i is the i th covariate; X denotes the mean o x i ; and X 1 denotes a value o increasing x i by 10% rom its mean value. This equation states how many percentage increase (or decrease) in the probability o entry occurrence will be obtained rom a 10% increase in the i th covariate. Table 6 summarizes the estimation results about the magnitude o the covariates eect on the hazard ratio. They reveal that the most important determinant o Taiwanese investment into China is the relative wage rate. The probability o Taiwanese irm s entry increases by about 50% when relative wage rate raises 10%. As or the exchange rate variables, the exchange rate level seems to have the largest eect on the hazard o entry occurrence. A 10% depreciation o NTD against RMB tends to increase the probability o the occurrence o market-seeking FDI by 19% while it may decrease the probability o the occurrence o export-substituting FDI by 11%. By contrast, a 10% increase in exchange rate volatility tends to increase the probability o export-substituting FDI by 13% while it may decrease the probability o market- seeking FDI by 8%. Other covariates, (e.g., exchange rate trend, marketing intensity, proit rate, R&D intensity, irm s size, and capital labor ratio), only have small inluences on the entry decision. It also indicates that, compared to the case when FUND=0, the probability o a irm s entry is about.13 times greater i its parent company can und the investment. 5 See Box-Steensmeier and Jones (004, p.60). 5

27 To sum up, our empirical indings indicate that relative wage rates, the real exchange rate level and its trend as well as volatility have a signiicant impact on the timing o Taiwanese irms investment into China. In particular, the relationship between real exchange rate volatility and the timing o FDI varies with the motives o investing irms, which suggests that it is important to consider this act in investigating the determinants o oreign direct investment. 5. Conclusion This paper theoretically and empirically examines how exchange rate uncertainty inluences the timing o FDI. We develop an integrated model o FDI under uncertainty to illustrate the impact o exchange rate volatility on the FDI activity o an export-substituting irm versus a market-seeking irm. We show that while exchange rate uncertainty tends to delay the FDI activity o a market-seeking irm, it actually may accelerate the FDI activity o an export-substituting irm i the degree o risk aversion o the irm is high enough. In other words, the relationship between exchange rate uncertainty and FDI is generally indeterminate, depending on the FDI types. The rationale behind these results is that market-seeking FDI might increase the exposure o the irm s proits to exchange rate risk, while export-substituting FDI might reduce it. Our theoretical results can be viewed as a synthesis o many previous studies on this issue. Firm-level data on Taiwanese irm s outward FDI into China over the period between 1987 and 00 are employed to test the validity o our theoretical results. The empirical indings indicate that real exchange rate movements have had a signiicant impact on Taiwanese irms investment into China. In general, the empirical results are consistent with the prediction o the theory. Our results reveal that the relationship between exchange rate uncertainty and FDI is crucially dependent on the motives o the investing irms. Hence, it is 6

28 essential to consider this actor in an empirical model so that the testing results are ree rom aggregation bias. 7

29 Appendix 1: Proos o Lemmas Proo o Lemma 1 Using Equations (8a) and (8b), it can be shown that R L σ µ + σ σ =Φ σ a p ρ µ ρσ ( µ 0.5 σ ) + ρ µ σ ( α ) (A1) and ( ρ µ ) R L ρ µ ρ µ 0.5σ 4 =Ψ + ( 1+ α ) 1 µ a p ρσ + ( µ 0.5 σ ) ρ µ σ ρ µ (A) where R0 α( ρ µ )[ P ( P W ) ] Φ= > 0 ; W (1 + α) ( ρ µ σ ) R0 αγ [ P ( P W ) ] Ψ = > 0. Since we assume W (1 + α ) ρ µ σ > +, ρ µ 0.5σ ρ µ σ > 1 and 4( ρ µ ) > 1. Thus, R 0 L µ ap >. Regarding Equation (A1), since ρ µ <, σ σ ρ µ ρ µ σ RL σ ap > 0 i α µ + σ > ρσ + ( µ 0.5 σ ). Note that µ + σ α ρσ + ( µ 0.5 σ ) α = ( ρ µ σ ) + 4ρ ρσ + ( µ 0.5 σ ) 3ρ ρσ + ( µ 0.5 σ ) + ( µ σ )( µ 0.5 σ ) Since 3ρ > ( µ + 0.5σ ) and ρσ + µ σ > µ σ under the assumption ( 0.5 ) 0.5 ρ µ σ > +, thus µ + σ α >. Consequently, R σ a > 0. ρσ + ( µ 0.5 σ ) L p Proo o Lemma From Equation (1b), we have ( ρ µ ) R H ρ µ ρ µ 0.5σ 4 =Γ + ( β 1) 1 µ a p ρσ + ( µ 0.5 σ ) ρ µ σ ρ µ (A3) βγ ( P W ) where Γ= > 0. Similar to Equation (A), we have ( β 1) RH µ ap > 0. 8

30 Appendix : Data Description The level o the real exchange rate, R, is the average bilateral real exchange rate, expressed in unit o NTD per RMB. It is calculated with a nominal exchange rate o NTD to USD, and that o RMB to USD; it is delated with Taiwan s CPI and China s CPI, respectively. The data o CPI are compiled rom the database o Taiwan Economic Journal (TEJ) and nominal exchange rates are rom the Central Bank o China (Taiwan). The real relative wage index, WAGE, deined as the ratio o the real annual average wage o China over the real annual average wage o Taiwan, is compiled rom the database o TEJ. Two measures o trend and volatility o the real exchange rate are used. First, µ Tasy and σ Tasy are deined respectively as a modiied average and a modiied standard deviation o the monthly changes in the log o the real exchange rate over the past 4 months; that is T T σ Tasy, t = rt j 1 rt j 1 T 1 + j 1 T + = j= 1 1, 1 σ Tasy, t = +, T µ Tasy, t T rt j + 1 j = 1 where r = log R log R ; T = 4 ; is the space time interval, equal to 1 T. j j j 1 Second, a GARCH process is adopted to estimate the volatility. With data covering the period rom 1985:01 to 00:1, we conduct the Augmented Dickey Fuller (ADF) test. The test result rejects the null hypothesis o unit root or ln R t. The estimated GARCH model is as ollows: ln Rt = ln Rt ln Rt 1 = ut, ( 1.4) ht = u , t 1 h t 1 (18.00) (9.54) ( 4.) where ln R t is the irst dierence o the real exchange rate; and h t is the conditional variance o the error term u. The numbers in parentheses are t-statistics. Thus, µ σ GARCH are deined respectively as t GARCH and 1 T σgarch, t ht j + 1 T j = = T, µ GARCH, t = ut j+ 1. T j= 1 The monthly nominal exchange rates are compiled rom Central Bank o China (Taiwan) and CPI are compiled rom the database o TEJ. The data on the duration o a irm s FDI, sources o unds (FUND) and sales o its subsidiary are obtained rom Survey on Taiwanese Firms in Mainland China, 003~004, 9