Demand uncertainty and the Joint Dynamics of Exporters and Multinational Firms

Transcription

1 Demand uncertainty and the Joint Dynamics of Exporters and Multinational Firms Cheng Chen (University of Hong Kong) Tatsuro Senga (Queen Mary University of London) Chang Sun (Princeton University) Hongyong Zhang (RIETI) RIETI Seminar, Sep

2 Introduction

3 Introduction ˆ Firms face substantial uncertainty when entering foreign markets, e.g., imperfect information about their idiosyncratic demand ˆ How large/important is such uncertainty? Can firms resolve uncertainty by accumulating experience? ˆ This paper: ˆ Explore data on multinational firms sales forecasts ˆ New fact: demand uncertainty declines with market experience ˆ Quantify the role of uncertainty and learning for the joint dynamics of trade and FDI 2

4 Preview of data and empirical findings ˆ Japanese MNE data: each affiliate reports their projected sales next year ˆ Forecast error (FE): difference between the realized sales and projected sales FE = log [R t+1 /E t (R t+1 )] ˆ Use FE as a measure of firms uncertainty ˆ New empirical facts: ˆ FE declines with affiliates experience in the foreign market ˆ Initial FE is smaller for first-time entrants if their parent firms have exporting experience in the same region 3

5 Quantify the role of uncertainty and learning ˆ Quantitative model: exporting and FDI choices + dynamic learning model (Arkolakis et al., 15) ˆ Can replicate the two empirical facts about FE ˆ Qualitatively match dynamics of FE, sales growth and exits ˆ Counterfactual experiments: ˆ change in trade costs - dynamic interaction between trade and FDI ˆ change in uncertainty on the foreign markets (i.e., variance of noisy component of demand shock) 4

6 Related Literature ˆ Uncertainty and activities in foreign markets ˆ Policy or demand uncertainty: Handley and Limao (14, 15), Carballo (15) ˆ Learning about demand: Albornoz (12), Akhmetova and Mitaritonna (13), Cebreros (15), Timoshenko (15), Conconni et al. (16) ˆ Structurally back out firms expectations: Morales and Dickstein (16) New: measure firms expectation and provide direct evidence ˆ Exporter and MNE dynamics: Ruhl and Willis (16), Fitzgerald et al. (16), Gumpert et al. (16), Garetto et al. (16) New: quantify the role of learning using forecast data ˆ Micro- and macro-level uncertainty: Bloom (09), Bloom et al. (14), Asker et al. (14) 5

7 Facts about FE

8 Data ˆ Japanese firm-level datasets prepared by the Ministry of Economy, Trade and Industry, ˆ Basic Survey of Japanese Business Structure: all firms with 50+ employees and Y30 mil assets, provides export info, firms each year ˆ Basic Survey of Overseas Business Activities: overseas affiliates of Japanese MNEs, 3200 parents affiliates each year ˆ Merged data: 2300 parents affiliates each year ˆ Affiliates report their projected sales for the next fiscal year 6

9 Definition and distribution of FE ˆ We define forecast error as FE log t = log [R t+1 /E t (R t+1 )] ˆ Distribution of FE log t Density Forecast error (log deviation) 7

10 Empirical Fact 2: previous exporting reduces FE ˆ Previous work suggests exporting experience reduces uncertainty in FDI (Conconi et al., 16) ˆ Data and sample selection ˆ We only know the parent firms exports to regions (Asia, North America, Europe, Middle East, Africa, Latin America, Oceania) ˆ Examine first-time entrants into the host-country/region ˆ Focus on manufacturing parent firms and manufacturing or distributional-oriented affiliates (wholesaler + retailers) 9

11 Empirical Fact 2: previous exporting reduces FE Frequency Percent Total % of affiliates have previous exporting experience (higher than Germany and Norway, lower than Belgium) 10

12 Empirical Fact 2: previous exporting reduces FE Dependent Variable: FE log (1) (2) (3) (4) Exp 1 > (0.065) Exp 1 > 0 or Exp 2 > (0.064) Exp Expe. > (0.070) Exp Expe (0.006) Industry FE Yes Yes Yes Yes Country-year FE Yes Yes Yes Yes N R Standard errors are clustered at parent firm level, * 0.10 ** 0.05 ***

13 Empirical Fact 2: previous exporting and FE - age profile Coef. on age dummies Affiliate Age (when projecting sales) Exp-1>0 95% CI Exp-1=0 12

14 Model and Quantification

15 Jovanovic (82) with exporting and FDI choices ˆ Re-interpret uncertainty about productivity in Jovanovic (82) as uncertainty about demand (a la Arkolakis et al. 15). ˆ Firms do not know their time-invariant demand shifter θ. Each period they observe a noisy signal a = θ + ε and they use a to update θ. ˆ Firms choose whether to serve the foreign market via export or via FDI. Exporting has higher variable costs and higher per-period fixed costs but lower entry costs, and provides the same amount of information about θ each period. 14

16 Setup ˆ Time is discrete ˆ Two-layer consumer demand U t = ( χ 1 δ i i Q δ 1 δ it where a t (ω) follows ) δ δ 1 (, Q it = e at (ω) σ ω Σ it a t (ω) = θ (ω) + ɛ t (ω), ɛ t (ω) i.i.d. N ˆ Demand for a particular Japanese variety ) σ q t (ω) σ 1 σ 1 σ dω. ( ) 0, σɛ 2 q t (ω) = Ỹt P t 1 δ χ jp P σ δ jp,t ea t (ω) p t (ω) σ, ˆ P t is the aggregate price index for all goods ˆ P jp,t is the ideal price index for Japanese goods 15

17 Setup ˆ Firms can serve the foreign market via export or via FDI ˆ trade costs (τ, f x, fx e ), FDI costs (f m, fm) e ˆ θ (ω) follows N ( θ, σθ 2 ) ( ) ˆ productivity follows log ϕ N ϕ, σϕ 2 ˆ Produces with only labor q t = ϕl t. 16

18 Timing 1. Exogenous mass 1 of entrants draw ϕ and θ, but θ is unknown 2. Entrants and incumbents: 2.1 entry of entrants (endogenous entry) 2.2 receive exogenous death shock with prob η 2.3 decide whether to exit, becoming exporter or becoming MNE (endogenous mode switching and exit) 2.4 choose employment l thus q 2.5 observe a and set price p to clear the market, update belief about θ 17

19 Belief updating After the firm observes a 1, a 2,..., a n 1, the posterior about θ is normal with mean µ n 1 and variance σ 2 n 1 µ n 1 = σ 2 ɛ σ 2 ɛ + (n 1) σ 2 θ θ + (n 1) σ2 θ σɛ 2 + (n 1) σθ 2 ā n 1 ; (1) where σn 1 2 = σɛ 2 σθ 2 σɛ 2 + (n 1) σθ 2. (2) ā n 1 1 n 1 n 1 i=1 a i for n 2; ā

20 Implications of the model ˆ Firms cannot prefectly foresee their sales because ˆ they are uncertain about θ ˆ they receive ε shock each period ˆ Learning about θ reduces firms forecast error over time ˆ σθ 2 and σ2 ɛ jointly determine FE for firms with little experience ˆ σɛ 2 determine FE for experienced firms - directly calibrated ( ) ( ) Var FE log θ µn 1 + ε n 1 = Var σ ( ) θ µn 1 = Var + σ2 ɛ σ σ 2 σ2 ɛ σ 2. 19

21 Calibration ˆ Normalization ˆ wage in the foreign country w = 1 ˆ wage in Japan w = 1 and ˆ total expenditure on Japanese goods Y = 1 ˆ mean of θ distribution is normalized to zero. ˆ mass of entrants J=1 ˆ export entry cost fx e = 0 (no comparison between exporting and domestic production in model) ˆ parameters calibrated without solving the model ˆ parameters calibrated by solving the model and matching moments 20

22 Calibration - without solving model Parameters Description Value Source σ Elasticity of substitution between Japanese goods 4 Bernard et al. (2003) δ Armington elasticity between 2 goods from different countries β Discount factor % real interest rate η Exogenous death rate 0.03 Average exit rates of multinational affiliates f m FDI per-period fixed costs 0 Flat profile of affilates exit rate over their life cycles τ Iceberg trade costs 1.3 Ghironi and Melitz (2005) σ ɛ Standard deviation of the transitory demand shock 1.05 Variance of forecast errors for affiliates that are at least 10 years old 21

23 Calibration - match moments Parameters Value Description Moments Data Model f x 0.05 export fixed cost average exit rate of exporters fm e 2.73 FDI entry cost fraction of exporters among active firms σ ϕ 0.45 std. dev. of log productivity share of exports in total sales of Japanese goods abroad σ θ 0.77 std. dev. of demand shifter fraction of experienced MNEs at age

24 Untargeted Moments ˆ FE-age profiles for MNE with and without exporting experience Age effects on FE in the data Average FE in the model Exp-1>0 Exp-1= Affiliate Age (when projecting sales) Exp-1>0 Exp-1= Affiliate Age (when projecting sales) 23

25 Untargeted Moments ˆ Sales-age profiles for MNEs and exporters Average log export Average log sales data model data model Parent-market age Affiliate Age 24

26 Untargeted Moments ˆ Sales-age profiles for MNEs and exporters Exporter exit rate data model Parent-market age 25

27 Overview of counterfactuals ˆ Change uncertainty σ 2 θ or σ2 ɛ ˆ FE differs across countries and helps to identify σ 2 θ and σ2 ɛ ˆ two sources of uncertainty have different implications on trade and FDI patterns ˆ Change trade costs τ and σ 2 ɛ at the same time to illustrate complementarity between trade and FDI 26

28 Heterogeneity in σ FE for young and old firms age >= BEL MYS NLD BRA ESP USA ITA FRA ARE DEU CAN KOR TWN IDN PHL IND AUS THAGBR CHN MEX VNM age = 1 RUS 27

29 Counterfactual: change σ θ (a) prob of entrants mode switching (b) mass of exporters, age = 7-9 base eqm 0.2 to exp to exp or fdi (c) fraction of experienced MNEs 5 (d) avg market expe for new MNEs experienced all 28

30 Counterfactual: change trade costs σ ɛ (a) prob of entrants mode switching (b) mass of exporters, age = 7-9 base eqm 0.2 to exp to exp or fdi (c) fraction of experienced MNEs 6 (d) avg market expe for new MNEs experienced all 29

31 Counterfactual: FDI elasticity wrt τ under different σ ɛ Table 1: Responses to trade liberalization under different values of σ ɛ σ ɛ = 1.05 σ ɛ = 1.55 change in τ log(mp sales) log(exports) log(exports/mp sales)