RISK AND LAND REALLOCATION IN VIETNAM AFTER THE DECOLLECTIVIZATION

Transcription

1 RISK AND LAND REALLOCATION IN VIETNAM AFTER THE DECOLLECTIVIZATION Clément Imbert & Yanos Zylberberg Oxford University, CREI (Universitat Pompeu Fabra) 1/ 24

2 Risk is risky Agricultural production in developing countries is a risky activity. Insurance markets are incomplete, especially for covariate risk. Exposure to covariate risk should affect production decisions. 2/ 24

3 Intuition 1. Start with an allocation of land which is not optimal. 2. Households will trade land to optimize their production. How does exposure to risk affect this reallocation? 3/ 24

4 Literature Theoretical literature is on risk management (Sandmo 1971 Rosenzweig and Binswanger 1993): suboptimal investment in risky activity (over) diversification of activities Empirical literature is on risk coping strategies: Savings (Paxson 1992), Labor (Kochar 1999), transfers (Fafchamps and Lund 2003) We know more about risk coping than about risk management (Fafchamps 2009) 4/ 24

5 Preview What we do: Study agriculture in Vietnam in : after the decollectivization. Characterize the reallocation of land between Estimate the effect of risk on land reallocation. What we see: High level of reallocation less in risky places. 5/ 24

6 Outline 1. Context 2. Theoretical Framework 3. Empirical Strategy 4. Results 6/ 24

7 Vietnam land reforms ( ) 1988 Vietnam s land reform (Resolution 10) abandons the collective farming system gives individual households long-term use rights over the collectives land and other resources Land Law extends land tenure to 20 years for annual crops and 50 years for perennial crops. permits exchange, transfer, lease, inheritance, and mortgaging of land-use rights. 7/ 24

8 Type of land ownership Allocated 50% Contract 4% Long Term Use 21% Long Term Use 88% Owned 23% Other 2% Auctioned 3% Auctioned 2% Rented in 4% Rented in 4% Sold 2% Sold 8% % of Land Cultivated for Annual Crops. Source: VHLSS / 24

9 Floods Source: Dartmouth Observatory of Floods 9/ 24

10 Typhoons Source: Joint Typhoon Warning Center (US Navy) 10/ 24

11 Setup Environment: N agents in a closed village, T land Technology: one final output produced with land (inelastic labor supply): F i(t i) = θ it α i Markets: Output market with given price 1. Land market is perfect with price p. Assume no savings and no transfers. Timing: 1st period: agents are endowed with T and produce (no choice there). N 2nd period: agents trade land and produce (optimization). 11/ 24

12 First period allocation In the first period, all farmers have T N each: F i(t i) = θ i( T N )α i Marginal productivity of land is not equal across agents: F i (T i) = θ iα( T N )α 1 12/ 24

13 Second period allocation Each producer i solves: max T i π i = θ it α i pt i Marginal productivity of land is equal across agents: Optimal landholding: T i θ i αt α 1 i Market clearing implies that i Ti = T Equilibrium price is p = [ (αθ i ) T = p = [ αθ i ] 1 1 α p 1 1 α ] 1 α 13/ 24

14 Second period allocation with risk We introduce an aggregate shock: F(T i) = A sθ it α i Each producer i now solves: with A s stochastic. First order condition yields Optimal landholding max T i E[u(c s)] = E[u(A sθ it α i pt i)] E s[(a sθ iαt α 1 i p)u (c s)] = 0 T R i = [ αθ i p E s[a su (c s)] ] 1 1 α E s[u (c s)] 14/ 24

15 Data Vietnam Living Standard Survey Panel of households surveyed in 1993 and rural communes and 3367 households. Flood and Cyclone Data information: Flood from Dartmouth Observatory of Floods (NASA). Cyclone from Joint Typhoon Warning Center (US Navy). 15/ 24

16 Descriptive statistics in 1993 Mean Median Total Output (1,000 Dongs) Agriculture Share 75% 100% Annual Crop Share 87% 97% Rice Share 73% 83% HH with No Output 2% Liquid Assets Flood exposure Cyclone exposure Source: Rural households in VHLSS / 24

17 Yield dispersion over time 17/ 24

18 Estimation (1) Optimal landholding: ln(t i ) = 1 1 α ln( α p ) α ln(θi). For any period t: ln(θ i) = ln(y it) αln(t it) + ν it. Convergence towards the optimum at rate ρ: We estimate β = ln(t it) = ρ[ln(ti ) ln(t it)] + ζ it = ρ 1 α ln(yit) ρ 1 α ln(tit) ρ 1 α in: + ρ 1 α ln( α p ) + ρ νit + ζit 1 α ln(t it) = βln( Yit T it ) + η c + ε it 18/ 24

19 Estimation (2) With risk, optimal landholding: ln(t R i ) = 1 1 α [ln( α p ) + ln(θi) + Es[Asu (c s)] E s[u (c s)] ]. E s[a su (c s)] E s[u (c s)] = φ i(θ i, R) where R is a measure of risk on A s. We measure exposure to risk in the commune Risk c and estimate β and γ in: ln(t it) = βln( Yit T it ) + γln( Yit T it ) Risk c + η c + ε it 19/ 24

20 Results - first-period allocation VARIABLE Land allocation 1993 Land allocation 1998 HH Size (0.010) (0.009) Prime Age (0.076) (0.084) Avg Educ (0.008) (0.007) Head Male (0.03) (0.034) Head Age (0.006) (0.009) Ethnicity (0.110) (0.081) Head Local (0.049) (0.048) Gov. Job (0.070) (0.047) Observations 2,943 2,786 R-squared Commune FE Yes Yes 20/ 24

21 Results - convergence toward optimal VARIABLE Change in Log Annual Crop Land Prod./Land 0.329*** 0.331*** (0.0246) (0.0752) Cons./Land 0.437*** 0.451*** (0.0423) (0.0483) Observations 2,726 2,645 2,732 2,651 R-squared Commune FE Yes Yes Yes Yes HH Controls No Yes No Yes 21/ 24

22 Results - convergence toward optimal (with risk) VARIABLE Change in Log Annual Crop Land Floods Typhoons Prod./Land 0.514*** 0.365*** (0.141) (0.113) Prod./Land*Risk * (0.239) (0.547) Cons./Land 0.551*** 0.526*** (0.0818) (0.0708) Cons./Land*Risk * * (0.125) (0.271) Observations 2,645 2,651 2,645 2,651 R-squared Commune FE Yes Yes Yes Yes HH Controls Yes Yes Yes Yes 22/ 24

23 To do list Empirical challenges: 1993 land allocation is not entirely explained by observables: what are the other factors? Exposure to risk is very spatially correlated: risky communes have different characteristics. Theoretical (and empirical) challenges: Besides annual land, farmers grow perennial crops (15% of crop output) and non agricultural business: riskier activities? profitable? Firesales, liquidity constraints? 23/ 24

24 Thanks 24/ 24

25 References References - appendix 25/ 24