WPS6065 Policy Research Working Paper 6065 Long-term Impacts of Global Food Crisis on Production Decisions Evidence from Farm Investments in Indonesia Manabu Nose Futoshi Yamauchi The World Bank Multilateral Investment Guarantee Agency Economics and Policy Group May 2012 Policy Research Working Paper 6065 Abstract Did the rise in food prices have a long-term impact on response to price increases, wealthy farmers invested more agricultural production? Using household-level panel in productive assets, while poor farmers increased their data from seven provinces of Indonesia, this paper finds financial savings as well as consumption. Price spikes that the price shock created a forward-looking incentive relax liquidity constraints, which increases investments to invest, which can dynamically enhance productivity among the richer while do so savings and consumptions in agriculture. It also finds that the impact of the price among the poor, possibly leading to diverging income shock on investment behavior differs by initial wealth. In inequality in the long run. This paper is a product of the Economics and Policy Group, Multilateral Investment Guarantee Agency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at mnose@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Long-term Impacts of Global Food Crisis on Production Decisions: Evidence from Farm Investments in Indonesia 1 Manabu Nose 2 and Futoshi Yamauchi 3 World Bank Key Words: Investment effect, Expectation, Anticipated shock, Liquidity constraint Sector Boards: POV, SOCPT, ARD 1 We thank Luc Christiaensen, Quy-Toan Do, Emanuela Galasso, Ivanic Maros, Owen Ozier, Rachel Heath, and Reno Dewina, Sony Sumaryanto and seminar participants at the World Bank for their useful comments. We are grateful to the Japan International Cooperation Agency for financial support and the Indonesian Center for Agriculture Socio Economic and Policy Studies (ICASEPS) for collaboration. We thank Sony Sumaryanto for sharing agricultural commodity price data, and Dini Inayati for collecting additional price data. The findings, interpretations and conclusions expressed in this paper are entirely those of the authors, and do not necessarily represent the views of the World Bank, its Executive Directors, or the countries they represent. 2 Corresponding author: Manabu Nose, The World Bank, 1818 H Street, NW, Washington DC, 20433, USA; Email: mnose@worldbank.org; Tel: +1-(202)-294-5370 3 The World Bank, 1818 H Street, NW, Washington DC, 20433, USA; Email: fyamauchi@worldbank.org 1. Introduction Between 2007 and 2008, world prices of food commodities rose dramatically. The global food price inflation was transmitted to the domestic market of Indonesia, where the agricultural sector’s share is about 19% of GDP and 41% of total employment. The rising food prices raised fears that the spike in food expenditures could worsen households’ well-being, especially among the poor. Welfare impacts of food price inflation can be particularly large in Indonesia since the average family spends about a half of its income on food 4. On the other hand, higher food prices increased agricultural profits and created large income gains to agricultural households compared to non-agricultural households (Ravallion, 1990; Yamauchi and Dewina, 2011). Interestingly, World Bank (2011) shows evidence that the positive impacts on producers seemed to outweigh the negative effects on consumers’ welfare in Indonesia. In the literature, a large number of papers have investigated the short-term impact of food price shock on poverty and welfare in different contexts (e.g. Ivanic and Martin (2008); Ravallion (1990) for Bangladesh; Ferreira, Fruttero, Leite, and Lucchetti (2011) for Brazil; Vu and Glewwe (2011) for Vietnam; Friedman, Hong, and Hou (2011) for Pakistan). However, its long-term effects on agricultural production are still yet to be explored and understood 5. If the rise in food prices creates positive net gains to farmers, it is important to understand how farmers utilize the gains and change their agricultural production activities. If farmers perceive that the 4 Paxson (1993) showed from Thailand that price changes significantly cause consumption fluctuation. Using the same sample from Indonesia, Yamauchi (2012a) also recently showed that recurrent seasonality of rice prices fluctuates birth weight, resulting in variations in subsequent child growth and schooling investments. 5 In recent years, there is an emerging academic interest in examining the long-term impact of political or economic shocks (Collins and Margo (2007) for the impact of the 1960s riots in American cities on the property value; Hornbeck (2012) for the impact of the 1930’s American Dust Bowl on the population and agricultural production also in the U.S). However, there is no research which investigates the long-term impact of food price crisis. 2 price change is rather permanent and/or if the income gain creates sufficient liquidity, it might give farmers an incentive to invest in production assets. On the other hand, if they perceive that the price shock is transitory, they will increase their savings for the future price fall, leaving their investment unaffected (Paxson, 1992; Rosenzweig and Wolpin, 1993; Kazianga and Udry, 2006). Which of the two effects dominates is ambiguous. Did the positive food price shock increase investments in production capitals, creating a dynamic positive impact on agricultural productivity? Or, did farmers increase their savings (e.g., to cope with future negative income shocks) or production inputs (e.g., fertilizer and labor) to temporarily increase their outputs? This question is related to “induced innovation� hypothesis which suggests that agricultural development is directed by the change in the conditions of factor and product markets (Hayami and Ruttan, 1985). In our context, if the food price spike increases real wage of hired labors, it might induce farmers to invest in productive assets to save the labor cost. Farmers may also substitute labors with fertilizer as the fertilizer price fell relative to agricultural real wages. As the output price of food items increases, farmers may also have stronger incentive to reallocate their resources from non-farming activities to farm investments in response to the price spike. 6 In this paper, we aim to answer the questions of (a) whether farmers increased farm investments to expand the production frontier and (b) whether any constraints bind farmers’ optimal strategies. 6 Using household panel data from Indonesia, Yamauchi (2012b) showed that (i) large farmers tend to acquire more lands by renting in land when non-agricultural real wages increase, and (ii) they install machines if agricultural real wages increase. In contrast, small farmers seemed not to change their behavior. Since land size is significantly larger in non-Java islands than Java, the above observations imply regional differences in dynamic patterns of landholding distribution along with increasing real wages. 3 We use two rounds of household panel surveys conducted in 2007 and 2010, which represent the main agro-climatic zones in Indonesia. Since the first round was prior to the 2007 food price crisis, followed after three years by the second round in 2010, we can examine the effect of the food price shock on investment decisions. Using (i) monthly provincial-level food price data and (ii) farm and plot-level data on farming activities, we construct a household-level price shock variable by weighting the commodity-specific price changes by the household’s production share. This measure captures farm-specific exposures to the food price crisis. First, we use the price shock variable to estimate the responsiveness of the producer’s investment to the price increase. Since the future price level is uncertain and is highly unpredictable to households, we also examine whether the expectation formation affects their forward-looking investment behavior, by decomposing the price shock into anticipated and unanticipated components. Second, we test whether farmers’ responses to the price shock were affected by household characteristics and their asset holdings including farm land. For example, if liquidity constraint is binding farmers, they cannot invest enough in productive assets even though they surely expect a higher price in the near future (thus, the expected returns to investments are high). Family labor endowment can also affect investment decisions if increased production responding to an increase of food price is well accommodated by available family labor, reducing necessity for additional physical investments. In the empirical analysis, we found that (i) farmers benefited from the positive price shock in general, but the anticipated components of the shock created an incentive to invest in productive assets. Their investment decision was, however, influenced by the initial wealth endowment, showing a clear contrast between wealthier and poorer farmers in the presence of food price increase. That is, poorer farmers tend to increase investments in productive assets, probably 4 because an increased food price relaxed their liquidity constraint disproportionately more for the poor. The result confirms that the investment decision depends on the conditional expectations of future price changes. Farmers anticipate the future price, which drives their investment decisions. Our analysis highlights the critical role of the initial wealth endowment in determining the trajectory of long-term agricultural development. The finding that poor farmers are likely to invest more in response to the price spike implies that inequality between large and small farmers can shrink, and small farmers’ investments in productive assets may narrow the existing gap in production capacity between them. Another key finding is that farmers change their investments in response to anticipated price changes, so the expectation formation plays an important role. Unanticipated parts of the price shock did not affect their investment behavior. The above finding is consistent with our theoretical prediction too. This paper is organized as follows. The next section set ups a simple model to clarify our intuitions. Section 3 explains our survey and provides background information. Sections 4 to 6 carry out empirical analyses. Finally, Section 7 presents conclusions. 2. A Simple Model 2.1 Environment In this section, we describe our theoretical framework. Consider a producer using land and capital as factors of production. In the first period, the producer can decide the investment for 5 the next period. For simplicity, we assume out capital depreciation. The producer can borrow and lend with an interest rate competitively determined in the credit market. Let 𝐴 denote landholding, and 𝑓(𝑘 ) represent land productivity. 𝑘 is per-land capital stock. The producer has income from agricultural production �𝐴𝑓(𝑘 ) where � is the output price. The producer lives in 2 periods. Budget constraints are: �1 = �1 𝐴𝑓(𝑘1 ) − Δ𝑘𝐴 + �  and �2 = �2 𝐴𝑓(𝑘2 ) − (1 + 𝑟)�. Borrowing � is allowed with gross interest rate 1 + 𝑟. If the producer saves (� < 0), there are positive returns in the next period (1 + 𝑟)� . With investment Δ𝐾 = 𝐴Δ𝑘, the next period capital stock is determined as                𝐾2 = 𝐴𝑘2 = 𝐴(𝑘1 + Δ𝑘)          (1) The producer maximizes the discounted sum of current and future utilities over Δ𝑘 and � : max{Δ𝑘,�} 𝑢 (�1 ) + 𝛽𝐸𝑣(�2 ), subject to the budget constraints and Eq. (1). At this stage, we do not impose any constraints on �. The standard Euler equations are:  𝑘2 :          𝑢′ (�1 ) = 𝛽𝑓 ′ (𝑘2 )𝐸[𝑣 ′ (�2 )�2 | Ω1 ]   (2a)  �:             𝑢′ (�1 ) = 𝛽(1 + 𝑟)𝐸[𝑣 ′ (�2 )| Ω1 ]         (2b) Since 𝐸[𝑣 ′ (�2 )�2 | Ω1 ]   = 𝐸 [𝑣 ′ (�2 )|Ω1 ]𝐸[�2 |Ω1 ] + 𝐶𝑜𝑣(𝑣 ′ (�2 ), �2 |Ω1 ), we obtain from Eq. (2a) and Eq. (2b): 𝐶𝑜𝑣 (𝑣 ′ (�2 ), �2 |Ω1 )   1 + 𝑟 = 𝑓 ′ (𝑘2 ) �𝐸 [�2 |Ω1 ] + �     (3) 𝐸 [𝑣 ′ (�2 )|Ω1 ] ≦ 𝑓 ′ (𝑘2 )𝐸[�2 |Ω1 ], 6 where the last inequality holds since 𝐶𝑜𝑣(𝑣 ′ (�2 ), �2 |Ω1 ) ≦ 0. 𝐸 [𝑣 ′ (�2 )|Ω1 ] The equality holds when there is no uncertainty in �2 or the producer is risk neutral. The covariance also captures the effect through which capital investment increases the variance of the future production value. 2.2 Anticipated vs Unanticipated Shocks Next we consider the expectation formation. Our main interest is to clarify theoretical insights of the potential effects of a price spike on the investment, financial savings and consumption. Proposition 1 Effect of Transitory Price Shcoks: If the price dynamics is i.i.d., then 𝐸 [�2 |Ω1 ] = 𝐸 [�2 ] = �∗ and 1 + 𝑟 = �∗ 𝑓 ′ (𝑘2 ) . That is, a temporary price shock in �2 (if the producer believes so) does not induce investments since it does not increase the expected marginal productivity of capital in the next period (as the producer thinks �2 moves back to normal Proposition 2 Effect of Anticipated Price Shocks: If 𝐸 [�2 |Ω1 ] increases in response to an increase of �1 (i.e., 𝐸 [�2 |Ω1 ] > �∗ ), there are two 𝐶𝑜𝑣(𝑣 ′ (�2 ),�2 |Ω1 ) effects through: (i) 𝐸 [�2 |Ω1 ] and (ii) . The effect (i) is positive to investments, but 𝐸[𝑣 ′ (�2 )|Ω1 ] the effect (ii) is negative to investments (i.e., due to risk aversion). When the first effect dominates, the investment will increase. 7 We illustrate the above predictions directly related to 𝐸 [�2 |Ω1 ]. Suppose that there is a structural change in price dynamics at the initial period, so that the producer has to learn about the new price distribution (i.e., the producer uses the updated distribution in each period to form the expectations). For example, we assume that there are price signals between times 1 and 2 (e.g., observing international prices and monthly domestic prices), so that the producer can learn about the price distribution and form the expectations of �2 . Signals are given as �𝑠 = 𝜃 + 𝜀𝑠 where 𝜃 is the unknown mean in the new regime (𝜃 ≠ �∗ ) and 𝜀𝑠 is an i.i.d. price shock, both following normal distributions. The producer knows that the mean price in the new/current region differs from that of the previous regime. In the above setting, rational learning (Bayesian) gives 𝐸 [�2 |Ω1 ] = 𝜙𝑠 �𝑠 + (1 − 𝜙𝑠 )𝜇1 (4) 2 (ℎ ) 𝜎𝑠 2 and 𝜙𝑠 = 2 2 (ℎ ) where 𝜇1 is the prior mean right after time 1, 𝜎𝑠 (ℎ) is the prior variance, 𝜎𝑠 (ℎ)+𝜎𝜀 2 and 𝜎𝜀 (ℎ) is the noise variance. Let ℎ denote the set of structural factors that determined both 2 2 𝜎𝑠 (ℎ) and 𝜎𝜀 (ℎ). Assume that these variances are taken from the past experience before the structural change. In other words, only the mean has changed. In the empirical analysis, we linearly approximate Eq. (6) to estimate 𝐸 [�2 |Ω1 ] (the anticipated price) and also derive an unanticipated component, �2 − 𝐸[�2 |Ω1 ]. Prediction: Anticipated vs Unanticipated Prices: Investments respond to the anticipated price 𝐸[�2 |Ω1 ], and not to the unanticipated shock. �2 − 𝐸 [�2 |Ω1 ] 8 2.3 Liquidity constraint There are two modifications we consider. First, we incorporate a liquidity constraint (imperfect credit market) in the above model by assuming that the interest rate depends on 𝐴. Suppose that 𝑟 is constant if � < 0 but 𝑟(𝐴) if � > 0 where 𝑟 ′ (𝐴) ≦ 0. That is, they face the same interest rate when they save, but small farmers have to pay a higher interest rate than large farmers when they borrow. When they borrow, it is often the case that they offer collateral (e.g., land). If so, since the left-hand side of Eq. (3) increases for small farmers, the marginal effect of 𝐸[�2 |Ω1 ] on investments becomes smaller. Another way to incorporate liquidity constraint is to impose the condition that � ≤ 0. They can only save but cannot borrow. The modified Euler equation for � is given as:              𝑢′ (�1 ) = 𝛽(1 + 𝑟)𝐸[𝑣 ′ (�2 )| Ω1 ]   + 𝜆,               (4)′ 𝜆 ≥ 0. where 𝜆 is the Lagrangian multiplier associated with � ≤ 0. Then, the condition (3) is modified as 𝜆 ′ 𝐶𝑜𝑣 (𝑣 ′ (�2 ), �2 |Ω1 ) (1 + 𝑟) + = 𝑓 (𝑘2 ) �𝐸 [� |Ω 2 1 ] + �     (5) 𝛽𝐸[𝑣 ′ (�2 )|Ω1  ] 𝐸 [𝑣 ′ (�2 )|Ω1 ] Even in Eq. (5), our intuition remains the same: it potentially reduces the impact of 𝐸 [�2 |Ω1 ] on the investment, but we also observe asymmetry of the liquidity effect since the likelihood of facing the constraint (𝜆 > 0 or 𝜆 = 0) depends on the realization of �1 . 9 Proposition 3 Effect of Liquidity Constraint: In the presence of borrowing constraint (i.e., farmers cannot borrow), the marginal effect of 𝐸 [�2 |𝛺1 ] will decrease. In the following sections, we describe our data and empirical strategies to verify the above theoretical predictions. 3. Empirical strategy We use the IMDG data to test the implication of our theoretical model on the expectation formation and liquidity constraints. In our empirical model, we assume that farmers are price takers. Using the exogenous variations of food price shocks to households, we estimate the relationship between agricultural investment and food price using a first-differencing (FD) model. We define food price variable ∆𝑃 as a growth rate of food prices between 2007 and 2009. The price index is constructed as follows. We start from the provincial capital producer prices of five main food items: rice (c=1), maize (c=2), cassava (c=3), estate plantation crops (c=4), and horticulture crops (c=5). We use an aggregate price measure of estate plantation crops which include major crops such as coffee, cocoa, and coconuts. Our price measure of horticulture crops also includes both vegetables and fruits. Since many farmers produce multiple crops in our sample, the price of each food item is weighted by the revenue share of each main crop among total revenues in 2007 ( φ 2007 ) 7 , and the weighted prices are summed. This method 7 Revenue shares are the proportions of production revenues of rice, maize, cassava, estate crops, and horticulture crops. If farmers do not market the crops and thus only report the volume of crop production, we imputed the production revenue using the median price of each crop at the village level. 10 provides household-level variations in the exposure to food price shocks, enabling the identification of food price shock impact. 5 ∆Pij = ∑ φijc , 2007 ( pc , j , 2009 / pc , j , 2007 − 1) c =1 (6) For a household i in province j, we estimate: ∆ ln k ij = β1 ∆Pij + β 2 ∆Pij2 + β 3 L0 + β 4 A0 + β 5 X 2007 + D j + ∆ε ij (7) where ∆ln𝑘 is the log of gross investments in agricultural assets (in real term) between 2007 and 2010. Landholding size (production scale) 𝐿0 , the durable asset (initial wealth endowment) 𝐴0 , and a vector of household characteristics in 2007 (i.e., the age of the household head, the household’s average years of schooling, and household size) 𝑋2007 are included. Depreciation of agricultural assets is not captured. The standard error is clustered at the village level. In order to obtain causal effects of food price on investments, we use two-year panel data and difference out household-level fixed effects (time-invariant unobserved heterogeneity). We can difference out the risk aversion, which is the covariance term in the RHS of Eq. (5), and farmer’s abilities and tastes which could affect their crop choice. These unobserved factors may be correlated with the production (revenue) weight of each crop in 2007 ( φ 2007 ), which may bias our estimate of the effect of food price on the investment decision. 11 We have a concern that unobserved time-varying provincial factors, which affect the rise in food price, could also affect households’ production decision. Different provinces have different natural endowments of land and resource, and the agro-climatic characteristics (e.g. soil quality) are also different, which affects actual realization of the food price changes. For example, Java and other provinces have very different levels of natural endowments, agricultural technologies, and market integration. We also expect that, over time, each province might have taken different measures to rising food prices. To account for the bias from the time-varying unobserved provincial attributes, we include provincial dummies (𝐷𝑗 ) in Eq. (7). If the food price had a dynamic impact on agricultural investments, we expect to see 𝛽 > 0 in Eq. (7). On the other hand, if the price change only had short-term impacts on the production level, we expect that the price variable does not have significant effect on investments but will only increase agricultural inputs. 4.1. Anticipated and unanticipated price shocks To test Predictions 1 and 2 on the expectation formation, we use a similar empirical strategy as Paxon (1992) and Jacoby and Skoufias (1997) to distinguish between anticipated and unanticipated components of the price shock. In the theory, producers form the expectation of 2 𝐸 (�2 ) based on 𝛺1 , which includes prior mean 𝜇1 , price signal �𝑠 , and noise variance 𝜎𝜀 (ℎ). We assume that households have rational expectations concerning the future distribution of food prices, and their price expectation formation is based on the lagged price growth ∆𝑃𝑡−1 , a vector of initial crop shares in 2007 (rice, maize, cassava, estate crops, and horticulture crops) w2007, household characteristics 𝑋2007, and province dummies. The initial crop shares summarize the information set (containing price dynamics up to 2007, given household characteristics such as 12 their risk aversion). Using the monthly price series of each crop from 2004 to 2007, we also compute the persistence and volatility measures. After removing linear trends of monthly price series, we include the first-lag autocorrelation 𝐴𝑅𝑡−1 as the price persistence measure, the standard deviation 𝑆𝐷𝑡−1 as the volatility measure, and the price trend 𝑇𝑅𝑡−1. We also include the interaction terms between initial crop shares in 2007 and average years of schooling (H) to allow the heterogeneity in price expectation formation depending on households’ human capital. In the first stage, we regress ∆𝑃𝑡 on these variables as follows. ∆𝑃𝑖𝑗𝑡 = 𝛼1 ∆𝑃𝑖𝑗𝑡−1 + 𝛼2 w𝑖𝑗,2007 + 𝛼3 𝑋𝑖𝑗,2007 + 𝛼4 w𝑖𝑗,2007 𝐻𝑖𝑗,2007 + 𝛼5 𝐴𝑅𝑡−1 + 𝛼6 𝑆𝐷𝑡−1 + 𝛼7 𝑇𝑅𝑡−1 + 𝐷𝑗 + 𝑒𝑖𝑗𝑡 (8) The anticipated component is the projection of the price change based on information available � to households in 2007, which is 𝑃 � 𝑃 . We use the residual of the price shock 𝑃 � 𝑈𝑃 = 𝑃 − 𝑃 𝑃 to represent the unanticipated component of the price shock. � ∆𝑃 𝑃 �∆𝑃 𝚤𝚥𝑡 = 𝛼1 𝑖𝑗𝑡−1 + 𝛼 � 2 w𝑖𝑗,2007 + 𝛼3 𝑖𝑗 ,2007 + 𝛼 �𝑋 � 4 w𝑖𝑗,2007 𝐻𝑖𝑗,2007 + 𝛼 �𝐴𝑅 5 𝑡−1 �𝑆𝐷 + 𝛼6 �𝑇𝑅 𝑡−1 + 𝛼7 𝑡−1 + 𝐷𝑗 � ∆𝑃 𝑈𝑃 � 𝑃 𝚤𝚥 = 𝑃𝑖𝑗 − ∆𝑃 𝚤𝚥 (9) Using Eq. (7) and Eq. (9), we estimate 𝛽 ’s in the following regression. This regression also ̂ depends in part on the controls for � = {𝐿0 , 𝐴0 } and 𝑋2007. Since the second stage estimator 𝛽 13 � , we use the two-step bootstrap estimation to adjust standard errors for first-stage estimator 𝛼 generated regressors. 2 2 �𝑃 + 𝛽 ∆𝑃 ∆ln𝑘𝑖𝑗 = 𝛽1 ∆𝑃 � 𝑈𝑃 � 𝑃 � 𝑈𝑃 � �𝑃 × � + 𝛽 ∆𝑃 𝑈𝑃 𝚤𝚥 𝚤𝚥 � + 𝛽4 �∆𝑃 2 𝚤𝚥 + 𝛽3 �∆𝑃 𝚤𝚥 � +𝛽5 ∆𝑃 𝚤𝚥 𝑖𝑗 6 𝚤𝚥 × �𝑖𝑗 + 𝛽7 �𝑖𝑗 + 𝛽8 𝑋𝑖𝑗,2007 + 𝐷𝑗 + ∆𝜀𝑖𝑗 (10) To estimate Eq. (10), we consider the possibility that farmers’ expectations on the future price may differ depending on the marketing arrangement of the products in 2007. In Indonesia, there exist traditional contractual arrangements such as Tebasan and Ijon 8 between farmers and traders, by which the purchasing prices of crops that traders offer differs from market prices at the harvest time. Experiencing a smaller price risk through an informal contract with traders, farmers may perceive future prices differently. For this reason, we do not include households who used either the Tebasan or Ijon system in our analysis. 4. Data and context 4.1 Survey The data come from two rounds of household surveys conducted in rural areas of Indonesia in 2007 and 2010 for 98 villages in seven provinces: Lampung, Central Java, East Java, West Nusa Tenggara (NTB), South Sulawesi, North Sulawesi, and South Kalimantan. The locations of surveyed villages are shown in Figure 1. In 2010, we revisited all the 98 sample villages to re- interview sample households and their splits of the 2007 survey households. 8 Tebasan is a harvesting practice in which standing crops, mainly paddy and maize, are sold on area basis just before harvesting at prices close to normal market rates. In this way, farmers and traders avoid transaction costs. While Ijon is purchase of crops prior to the harvest at lower price, so that farmers can avoid harvest risks by paying risk premium to traders. 14 Over the three years, some household members split from the 2007 original households and became an independent family head (for marriage or other reasons). In our sample, household division occurred in 204 original households (9% of our sample). We use the 2007-survey original household as the unit for analysis to avoid bias that may arise from household splits. For instance, a new household head, who split from his original household, might share and cultivate farm lands with his parents though the land is still owned by his parents (vice versa). By aggregating original and split households in 2010, we minimize the split bias 9. The 2007 survey was designed to overlap with villages in the 1994/95 PATANAS survey conducted by ICASEPS to build household panel data. The 1994/95 PATANAS survey focused on agricultural production activities in 48 villages chosen from different agro-climatic zones in seven provinces. In 2007, we visited those villages to expand the scope of research as a general household survey under the IMDG survey. In the 2007 round, therefore, we added 51 new villages in the same seven provinces 10. In the revisited villages in 2007, we re-sampled 20 households per village from the 1994/95 sample and followed the split households. In the new villages, we sampled 24 households from two main hamlets in each village. Since one of the 48 villages in the 1994/95 PATANAS was 9 It is possible that the 2007-08 food price crisis affected household split decisions, which will potentially cause an additional bias if we omit split households or do not aggregate original and split households. In this paper, we only report the results using the aggregated households. However, even when we use only households which did not experience split (excluding 204 split households), our empirical results remain the same. 10 These new villages were selected with the following criteria. First we chose the same districts where PATANAS villages are located. We list villages that had received relatively large amounts of government infrastructure projects during the period of 1995 to 2005, funded by either the Japan Bank for International Cooperation or the World Bank. Finally, the new villages were randomly sampled from the list. 15 not accessible for safety reasons in the 2007 survey (in NTB province), we have the total of 98 villages (in 39 kabupatens and 48 kecamatans). 4.2 Descriptive evidence Agricultural production Using the survey data, summary statistics provide basic information on agricultural households producing rice, maize, cassava, estate plantation crop, and horticulture crop in 2007 (N=762). The agricultural sector in Indonesia is characterized by smaller size of land (mean is 1.4 Ha in 2007) and multiple-crop farming. Many farmers cultivated three or four crops in Central and East Java. There was a dramatic increase in agricultural production from 2007 to 2010, which seems to be accompanied by large investments in production capital (i.e. owning farm building; having machineries such as tractor, thresher, sprinkler, irrigation pump, sprayer, and dryer; buying agricultural tools; see Table 2 for a detailed breakdown of investments). Table 3 shows the regional variation of the average agricultural production revenues in 2007 for rice, maize, cassava, estate plantation crops, and horticulture crop from IMDG 2007. Based on food crop statistics by the Badan Pusat Statisti (BPS), Java is the major producing area of rice, maize, and cassava as well as farming multiple crops including various vegetables. In remote areas in South Kalimantan and NTB, farmers primarily produce rice, whereas farmers in Lampung, and North and South Sulawesi are specialized in estate plantation crops. 16 Household characteristics Besides agricultural activities, the household module data of the IMDG shows that durable asset holding also increased from 2007 to 2010. In the table, durable asset holding is adjusted for household size (i.e. real values per household member). We use the provincial level consumer price index (2007=100) available online on the BPS website to calculate real values. Also, we use trimmed data to remove outliers throughout the analysis of this paper. It is defined as the ownership of non-production assets (such as residential house and land; consumer electrical appliances such as TV, radio, satellite antenna, and telephone) and the value has increased by 12.3% from 2007 to 2010. Local food price data The wholesale price of agricultural products has started rising since 2004 and the price increased substantially after 2007. Figure 2 disaggregates the price dynamics into five crops based on the monthly price data available from the Indonesian Bureau of Logistics (Bulog) and BPS, which shows that prices of all crops increased quite fast between 2007 and 2010. The prices of rice, maize, and cassava were taken from Bulog statistics, while those of estate crops and horticulture crops are from BPS. For the monthly raw price data of each crop, the Phillips-Perron test statistic (with time trend) does not reject the null hypothesis of unit root at any critical values, confirming that all price series are non-stationary. Table 1 shows that the first-order autocorrelations and standard deviations using the de-trended version of the monthly price series (which removes linear trends for each province and year). In general, the persistence of food price decreased and the volatility increased from pre- to post-crisis period, which implies that the movements of food price have become more uncertain in the post-crisis regime. 17 In the dynamics of local prices in each province, two features attract our attentions. First, due to the regional differences in the dynamics of food price in the post-crisis regime, the regional dispersions of prices have widened since 2007. Second, the prices of maize and cassava increased discontinuously with steps after 2007. In sum, the price level, the persistence, and the volatility seem to differ due to the different level of agricultural market integration. 5. Empirical Results 5.1 Capital investment decisions In Table 4, we aim to empirically clarify the effects of price shock (including its non-linearity) and household characteristics on farm investments. The table shows benchmark specifications using the FD model of Eq. (7), which we also use with some modifications in other tables coming later. We use the sample of 762 agricultural households who made positive amounts of agricultural investments between 2007 and 2009. Note that, as we found the price effect was clearly concave in the preliminary analysis, all specifications include the square term of the price change. In column (1), we assume identical investment responses to price shocks across all farmers. The negative estimate of the square term of ∆𝑃 indicates that the impact of food price shocks on households’ investment decision was non-linear (concave), though the linear effect is insignificant. Both the initial wealth and the landholding size had significantly positive effects on agricultural investments. A large family size helps to expand agricultural activities by increasing investments, as they are endowed with larger family labor force. 18 In column (2), we allow heterogeneity in the price effect by including its interactions with the initial landholding size and wealth level. If land is used as a collateral, small farmers are likely to face borrowing constraint and, as a result, small farmers tend to invest less. It is also possible that large agricultural land owners have higher profitability per acre (scale economy) and the return to investment is scale-dependent. 11 Although we cannot perfectly separate borrowing constraint from the scale effect since land is a collateral as well as the most important factor determining the scale of production (thus, the efficiency of investments since, e.g., machines are more effectively used on large farms), we can test whether the size of land affected the effect of price shocks on the investment decision. The theory also predicts that rich farmers invest more than poor farmers since their borrowing cost is lower. On the one hand, if this prediction is true, poor households will absorb the positive income shock by increasing savings and consumption. However, it could also be the case that poor farmers have a stronger incentive to invest since the marginal return to capital is higher than for rich farmers. It all depends on which activity – investment or consumption (current or future) – was constrained. With higher agricultural profit caused by the positive price shock, poor households might take this chance to invest since they do not have to borrow with a high interest rate. To test which hypothesis holds in our context, we interact the price change and durable assets in 2007 (𝐴0 ), a measure of the initial wealth level.. This measure does not include agricultural land but include residential land. Both linear and square terms of land size are positively signed, which implies that the effect of agricultural land on investment is increasing with convexity. Therefore, regardless of price shocks, large farmers tend to invest more. However, the interaction term of the price shock 11 Foster and Rosenzweig (2011) found the positive relationship between farm size and productivity using panel data in India. 19 variable and farm land is insignificant, suggesting that the effect of price shocks on investment decisions does not depend on the size of farm land. In the same column, the positive coefficient of durable assets implies that wealthier farmers could increase their agricultural investments even without food price shocks. However, the negative coefficient of the interaction term between the price shock and the initial wealth shows that the marginal effect of the price shock is greater for poorer farmers. Poorer farmers responded more to the price shock by investing, as it created an extra income gain, mitigating liquidity constraints for poor households. The negative marginal effect supports the conjecture that poorer households, which are likely to be liquidity constrained, had a greater incentive to invest due to higher returns to capital than wealthier farmers. 5.2 Decomposition of price shock variable The above results indicate that farmers expected that an increased food price would sustain in the near future and therefore decided to increase their agricultural investments during the food price crisis. The marginal effect diminishes when the price shock became too high. To better understand the non-linear price effect on farm investments, we decompose ∆𝑃 in Eq. (7) into price changes from 2007 to 2008 (∆𝑃1 ) and from 2008 to 2009 (∆𝑃2 ). We also include the interaction term of ∆𝑃1 and ∆𝑃2 . We expect the diminishing return to price shocks (i.e., ∆𝑃1 × ∆𝑃2 is negatively signed) since both utility and production functions are strictly concave. In addition, if we reply on a simple framework of Bayesian learning (described in Section 2), farmers put a higher weight on price 2( ) signals �𝑠 at the initial stage (that is, learning is fast) since the prior variance 𝜎𝑠 ℎ is the 20 biggest in the beginning. As the learning speed is faster at the initial stage, farmers should respond more to initial price shocks ∆𝑃1 . On the other hand, if farmers are uncertain about future price changes and are concerned about the expected payoff in subsequent years, they can increase their investments after observing the realization of ∆𝑃2 . Figure 3 shows that farmers were exposed to larger positive price shocks in the initial stage compared to subsequent years. The estimates in Table 5 show that the initial price shock ∆𝑃1 had a larger impact on investment decisions than ∆𝑃2 , which supports our learning hypothesis as well as the diminishing returns. ∆𝑃1 × ∆𝑃2 is negatively signed, which indicates that the impact of price shocks on investments has diminished over time. The results on the land size and the initial wealth are consistent with the results in Table 4. In column (2), we found that linear and square terms of land size are positive, showing that the scale effect remains positive and convex. In addition, we found that large land owners significantly increased their investments in response to ∆𝑃1 by 44.3%. In terms of the initial wealth, richer farmers had a greater incentive to increase investments, but the marginal effect of the initial price shock ∆𝑃1 diminishes by 52.8% for wealthier farmers. These results suggest that a higher food price did create a forward-looking incentive to invest, which can enhance productivity in the long run. However, Propositions 1 and 2 in Section 2 predict that investment decisions might differ by the nature of shocks whether anticipated or unanticipated, if household’s expectations on price dynamics matter. If households perceived that the price shock is temporary (i.e. unanticipated shock), it will not have impacts on investments (from Proposition 1). On the other hand, if households changed their expectations based on a permanent change of the price distribution, we predict that the price shock will affect their investment decisions. In the next sub-section, we will examine whether farmers’ 21 expectation formation matters by distinguishing between anticipated and unanticipated price shocks. 5.3 Anticipated vs. Unanticipated shocks In Table 6, we aim to incorporate the price expectation formation after the food price shocks by decomposing price changes into their anticipated and unanticipated components using the first- stage regression of Eq. (8). Table 7 reports the two-step bootstrap estimates of equation (10), which allows the anticipated and unanticipated components to have separate coefficients. Similar to Tables 4 and 5, land size, the initial wealth, and household characteristics are also controlled. We use 737 farmers, excluding those households using Tebasan or Ijon system. In the first stage (Table 6), negative estimates of the lagged price growth ∆𝑃𝑡−1 and the pre- crisis trend 𝑇𝑅𝑡−1 imply that farmers anticipated a structural change in the future food price dynamics after 2007. As expected, the sign of the persistence measure 𝐴𝑅𝑡−1 is positive (significant at 10%) and the sign of the volatility measure 𝑆𝐷𝑡−1 is negative (significant at 1%). In Table 7, we found that the two types of price shocks have different impacts on farm investments. Column (1) of the table shows that only the anticipated component of the price shock has a significantly positive effect on productive investments, while the effect of unanticipated component is insignificant. Farmers decide to invest in productive assets in response to the predicted component of the price change. In column (2), we aim to check possible non-linearity in the effects of the two types of price shocks on farm investments. We find that that (a) the effect of the anticipated price shock on investments is concave as confirmed in the previous section, while the unanticipated shock had 22 no significant effect, (b) the marginal effect of the anticipated price shock is increasing as the land size becomes larger, and (c) the marginal effect of the anticipated price shock is decreasing as the initial wealth increases. Less wealthier farmers have a greater incentive to invest when they anticipate an increase in price (the interaction term is significant at 5%). The result indicates that households’ general wealth level (i.e. ownership of non-liquid wealth) is an important determinant of productive asset investments. For the anticipated shock, the marginal effect is peaked at ∆𝑃𝑃 =0.15 for poor farmers and at ∆𝑃𝑃 =0.13 for rich farmers. In both groups, the effect diminishes as the anticipated shock increases beyond the above peaked price levels. As shown in the lower panel, for large land owners, the positive price effect becomes larger linearly with the anticipated price shock. The impact of the unanticipated shock on farm investments is always negative, regardless of the initial wealth and land sizes). 12 Finally, in column (3), we examine whether households’ risk aversion on future price volatility reduced their incentives to increase farm investments. For this purpose, we adopt the method used in Kazianga and Udry (2006) and investigate whether higher moments of future price shocks affected investment decisions by controlling for a price volatility measure in Eq. (10). We assume that farmers rationally expected the volatility of the post-crisis food prices based on the difference between realized prices 𝑃𝑡 and the mean price level (𝐸𝑡−1 (𝑃𝑡 )), which they (we) can predict based on the pre-crisis food price data from 2004 to 2007. We include the 12 We note that the concavity of anticipated price shocks is partly driven by some outliers (i.e., those with anticipated price shocks above one), and therefore the concave relationship is not quite robust. However, even without the concave relationship, the impact of anticipated price shocks on farm investments will be peaked around the similar points even after trimming these outliers. 23 �𝑡 (re-scaled to 102 ), where 𝑆𝐷𝑡 is prediction error of the standard deviation, defined as 𝑆𝐷𝑡 − 𝑆𝐷 the realized standard deviation in the post-crisis regime. 2009𝑚12 1 �𝑡 = � 𝑆𝐷 � [𝑃𝑡 − 𝐸𝑡−1 (𝑃𝑡 )]2 36 𝑡=2007𝑚1 Figure 4 depicts the distribution of 𝑆𝐷�𝑟𝑒𝑑𝑖�𝑡𝑖𝑜𝑛 𝑒𝑟𝑟𝑜𝑟 . The median of the prediction error is - 0.424, which means that farmers predicted a larger price volatility than the realized level of price volatility. . There were, however, many farmers who experienced a larger price volatility than they expected. If the realized price is more volatile in the post-crisis regime than they expected, farmers should have a smaller incentive to increase investments in response to the food price spike and would increase pre-cautionary savings. The result in column (3) shows that 𝑆𝐷�𝑟𝑒𝑑𝑖�𝑡𝑖𝑜𝑛 𝑒𝑟𝑟𝑜𝑟 has a negative sign but insignificant. Even in this specification, the results on anticipated and unanticipated shocks remain robust as we found in column (1) and (2). This finding reinforces our arguments that not all farmers who anticipated a positive price shock invested in agricultural productive assets, but poorer farmers did. 6. Conclusion In this paper, we have examined farmers’ investment decisions during the food crisis period in Indonesia using recent household panel data. The empirical analysis showed the positive price impact on farmers’ investments responding to the food price increase experienced in the period 2007 - 2008. Unlike the negative welfare impacts of higher food prices for consumers studied in the previous literature, we found that the anticipated component of the price shock created a 24 forward-looking incentive for Indonesian farmers to invest in productive assets. The above effect was strong among poor farmers. There are some interesting implications. Whether the investment indeed had a dynamic positive impact on agricultural productivity is an important question. Our preliminary result suggests that investments undertaken between 2007 and 2010 have positively impacted productivity gains amid the global food price crisis. This implies that the food price spike seemed to induce an upward shift in the production frontier in agriculture. Moreover, whether, in the long run, the food price crisis led to a divergence between rich and poor farmers is an important question to answer. Our results indicate that, though large landholders tend to invest in productive assets regardless of price shocks, small farmers capture the price spike as a rare opportunity to relax their liquidity constraint and increased their investments. References Carter. M, Little. P, Mogues. T, and Negatu. W (2007) “Poverty Traps and Natural Disasters in Ethiopia and Honduras�, World Development, 35(5), pp. 835-856 Collins. W and Margo. R (2007) “The Economic Aftermath of the 1960s Riots in American Cities: Evidence from Property Values�, The Journal of Economic History, 67(4), pp. 849- 883 Deaton. A (1990) “Saving in Developing Countries: Theory and Review�, Proceedings of The World Bank Annual Conference on Development Economics, pp. 61-96 Dercon. S and Christiaensen. L (2010) “Consumption Risk, Technology Adoption, and Poverty Traps: Evidence from Ethiopia�, Journal of Development Economics, 96(2), pp. 159-173 25 Dercon. S and Krishnan. P (1998) “Changes in Poverty in Rural Ethiopia 1989-1995: Measurement, Robustness Tests and Decomposition�, University of Oxford WPS/98-7 Fafchamps. M and Pender. J (1997) “Precautionary Saving, Credit Constraints, and Irreversible Investment: Theory and Evidence from Semiarid India�, Journal of Business and Economic Statistics, 15(2), pp. 180-194 Ferreira. F, Fruttero. A, Leite. P, and Lucchetti. L (2011) “Rising Food Prices and Household Welfare: Evidence from Brazil in 2008�, World Bank Policy Research Working Paper No. 5652 Foster. A and Rosenzweig. M (2011) “Are Indian Farms Too Small? Mechanization, Agency costs, and Farm Efficiency�, mimeo, Brown University Friedman. J, Hong. S. Y., and Hou. X (2011) “The Impact of the Food Price Crisis on Consumption and caloric Availability in Pakistan: Evidence from Repeated Cross-sectional and Panel Data�, mimeo, World Bank Hayami, Y. and Ruttan, V. (1985) “Agricultural Development: An International Perspective�, Baltimore: Johns Hopkins University Press Hornbeck. R (2012) “The Enduring Impact of the American Dust Bowl: Short and Long-run Adjustments to Environmental Catastrophe�, American Economic Review, 102(4), pp. 1477- 1507 Ivanic. M and Martin. W (2008) “Implications of Higher Global Food Prices for Poverty in Low-income Countries�, Agricultural Economics, 39, pp. 405-416 Jacoby. H and Skoufias. E (1997) “Risk, Financial Markets, and Human Capital in a Developing Country�, Review of Economic Studies, 64(3), pp. 311-335 Kazianga. H and Udry. C (2006) “Consumption Smoothing? Livestock, Insurance and Drought in Rural Burkina Faso�, Journal of Development Economics, 79(2), pp. 413-446 26 Paxson. C (1991) “Using Weather Variability to Estimate the Response of Savings to Transitory Income in Thailand�, American Economic Review, 82(1), pp. 15-33 -----------, (1993) “Consumption and Income Seasonality in Thailand,� Journal of Political Economy, 101(1), pp. 39-72 Ravallion. M (1990) “Welfare Impacts of Food Price Changes under Induced Wage Responses: Theory and Evidence from Bangladesh�, Oxford Economic Papers, 42(3), pp. 574-585 Rosenzweig. M and Wolpin. K (1993) “Credit Market Constraints, Consumption Smoothing, and the Accumulation of Durable Production Assets in Low-income Countries: Investments in Bullocks in India�, Journal of Political Economy, 101(2), pp. 223-244 Townsend. R (1994) “Risk and Insurance in Village India�, Econometrica, 62(3), pp. 539-591 Udry. C (1995) “Risk and Saving in Northern Nigeria�, American Economic Review, 85(5), pp. 1287-1300 Vu. L and Glewwe. P (2011) “Impacts of Rising Food Prices on Poverty and Welfare in Vietnam�, Journal of Agricultural and Resource Economics, 36(1), pp. 14-27 World Bank (2011) “Boom, Bust and Up Again? Evolution, Drivers and impact of Commodity Prices: Implications for Indonesia�, Washington D.C., World Bank Yamauchi, F. (2012a) “Prenatal Seasonality, Child Growth and Schooling Investments: Evidence from Rural Indonesia,� Journal of Development Studies. 48(9), pp. 1323-1341 Yamauchi, F. (2012b) “Flexible Labor Market, Landholding and Mechanization in Agriculture: Evidence from Indonesia�, mimeo, International Food Policy Research Institute. Yamauchi. F and Dewina. R, (2012) “Risks and Spatial Connectivity : Evidence from Food Price Crisis in Rural Indonesia�, Food Policy, 37(4), pp. 383-389 Zeldes. S (1989) “Consumption and Liquidity Constraints: An Empirical Investigation�, Journal of Political Economy, 97(2), pp. 305-346 27 Summary statistics N Mean sd p50 min max d_ln(Inv) 762 12.54012 1.839045 12.4676 8.289341 19.43883 d_price 762 0.2261237 0.3029745 0.2136982 -0.226667 2.12 d_price1 762 0.2218339 0.2817534 0.1915153 -0.2 1.8 d_price2 762 0.0075981 0.0733628 0.0007402 -0.355556 0.3333333 Ln(asset) 762 16.91856 1.16201 17.05605 11.3362 22.79049 Ln(land) 762 -0.626789 1.250713 -0.525945 -5.809143 2.197225 Average years of schooling 762 7.569207 2.92664 7.225 0 16 Household size 762 5.328084 2.067195 5 1 16 Average age 762 35.29266 5.703282 35 20 55 TR_t-1 762 26.37879 27.18154 14.84064 -2.266692 80.62788 AR_t-1 762 0.6863148 0.1308561 0.7125431 0.2308874 0.8474569 SD_t-1 762 187.6423 183.419 111.8246 10.67901 718.7351 SD_prediction error 737 -0.9322844 1.294098 -0.4240843 -4.242996 1.612561 (Note) Durable asset holding is converted to real term, deflated by CPI index, and is adjusted for household size. 28 Table 1: Monthly Commodity Prices (2004-2010) 1. Autocorrelations (1 month lag) Banjarmasin Lampung Makassar Mataram Pre Post Pre Post Pre Post Pre Post Rice 0.844196 0.742783 0.623171 0.736379 0.650719 0.157271 0.444033 0.397506 Maize 0.821649 0.827385 0.116147 0.843246 0.8267 0.724469 0.557993 0.473394 Cassava 0.864643 0.76801 0.471296 0.476669 0.671444 0.76885 0.7887 0.807267 Estate 0.713411 0.731777 0.734365 0.792293 0.821507 0.850851 0.510722 0.585104 Horticulture 0.652521 0.510954 0.604389 0.741782 0.729974 0.25017 0.362384 0.547258 Average 0.779284 0.716182 0.509874 0.718074 0.740069 0.550322 0.532767 0.562106 Menado Semarang Surabaya All Pre Post Pre Post Pre Post Pre Post Rice 0.675637 0.584909 0.823902 0.349811 0.654806 0.525774 0.673781 0.499205 Maize 0.703467 0.612008 0.788834 0.702844 0.763673 0.850674 0.654066 0.719146 Cassava 0.772064 0.857352 0.847457 0.842138 0.75803 0.802342 0.739091 0.760375 Estate 0.712543 0.805274 0.756301 0.873499 0.841585 0.821223 0.727205 0.780003 Horticulture 0.607716 0.572798 0.675657 0.554181 0.687848 0.537905 0.617213 0.530721 Average 0.694286 0.686468 0.77843 0.664495 0.741188 0.707583 0.682271 0.65789 2. Standard deviations Banjarmasin Lampung Makassar Mataram Pre Post Pre Post Pre Post Pre Post Rice 718.7351 202.1569 290.385 363.8876 244.2571 95.09454 215.8059 167.7218 Maize 95.49091 354.9766 217.2418 729.6438 80.84213 533.2995 224.5837 127.4998 Cassava 88.14651 421.188 228.0822 146.6009 68.04749 107.7264 257.8521 260.2294 Estate 34.98767 35.93951 108.3112 55.1585 111.552 82.88177 10.67901 5.14049 Horticulture 33.01025 2.011518 23.59837 4.562912 17.42777 1.982555 16.90491 6.173323 Average 194.0741 203.2545 173.5237 259.9707 104.4253 164.197 145.1651 113.353 Menado Semarang Surabaya All Pre Post Pre Post Pre Post Pre Post Rice 257.1523 197.8883 249.6365 86.9732 190.6296 178.155 309.5145 184.5539 Maize 147.2047 264.1175 85.06493 173.7563 147.8522 240.0603 142.6115 346.1934 Cassava 129.9688 579.6494 95.79269 656.522 90.74894 252.6408 136.9484 346.3653 Estate 152.8929 69.11196 21.53614 31.76717 45.2243 54.8653 69.31189 47.83781 Horticulture 28.72062 5.073268 20.00913 2.8141 22.69024 4.82508 23.19447 3.920394 Average 143.1879 223.1681 94.40788 190.3666 99.42906 146.1093 136.3161 185.7742 Pre = Pre-food crisis (from 2004m1-2006m12) Post = Post-food crisis (from 2007m1-2010m12) (Note) Autocorrelations and standard deviations are computed based on the de-trended series of each crop and province using pre- and post-crisis monthly price data. We remove linear time trend to construct de-trended price series. Average is the simple average. (Source) Bulog, BPS 29 Table 2: Type of investment in agricultural asset N Mean (1000Rp) Min Max Farm equipment* 1,110 990.8 2.5 320,000 Sprayer 404 707.8 0 100,000 Farm building 223 3,294.3 50 120,000 Barn 43 482 0 9,900 Plow/Harrow 30 538 30 3,000 Tractor 21 8,561.9 100 20,000 Dryer 20 141.3 0 2,600 Machinery** 19 7,744.7 250 30,000 Thresher 17 9,700 0 70,000 *Farm equipment include agricultural tools such as shovel, hoe, sickle, jackknife, machete, crowbar and so on. **Milling, feed processor, crumb rubber processor etc. Table 3: Average share of production revenues (by provinces) All Each provinces provinces Central East Java Lampung North South South NTB Java Sulawesi Sulawesi Kalimantan Rice 0.369 0.128 0.215 0.283 0.317 0.350 0.596 0.593 Maize 0.060 0.071 0.185 0.011 0.050 0.114 0.040 0.024 Cassava 0.047 0.099 0.024 0.133 0 0.008 0.015 0 Estate crops 0.336 0.151 0.108 0.549 0.598 0.489 0.261 0.108 Horticulture crops 0.189 0.550 0.275 0.024 0.035 0.039 0.088 0.275 30 Table 4: Positive price shock and investment decisions (1) (2) (3) VARIABLES d_lninv d_lninv d_lninv d_price -0.425 7.218** 7.148** (0.748) (3.112) (3.252) d_price^2 -0.567* -0.582* -0.529* (0.301) (0.306) (0.306) d_price*Ln(asset) -0.445*** -0.446** (0.167) (0.175) Ln(asset) 0.179** 0.284*** 0.280*** (0.0756) (0.0927) (0.0935) d_price*Ln(land) 0.364 0.400 (0.246) (0.254) Ln(land) 0.222*** 0.152* 0.276*** (0.0570) (0.0810) (0.0947) Ln(land)^2 0.0542*** (0.0175) Average years of schooling 0.0128 0.0150 0.0155 (0.0272) (0.0270) (0.0268) Household size 0.0878** 0.0841** 0.0759** (0.0334) (0.0342) (0.0347) Average age -0.00242 -0.00311 -0.00330 (0.0107) (0.0107) (0.0107) Constant 9.124*** 7.346*** 7.431*** (1.247) (1.518) (1.519) Observations 762 762 762 Sample All All All Provincial dummies included but not reported YES YES YES R-squared 0.223 0.233 0.239 *** p<0.01, ** p<0.05, * p<0.1 Standard errors are heteroscedasticity robust and clustered at the village level, and are reported in the parentheses. 31 Table 5: Decomposed price shock and investment decisions (1) (2) VARIABLES d_lninv d_lninv d_price1 8.713*** 8.612** (3.267) (3.351) d_price2 5.702 5.100 (14.63) (14.11) d_price1*d_price2 -14.19** -13.30** (6.296) (6.372) d_price1^2 -0.106 -0.117 (0.492) (0.502) d_price2^2 -14.80 -15.03 (12.39) (12.39) d_price1*Ln(asset) -0.532*** -0.528*** (0.178) (0.183) d_price2*Ln(asset) 0.0302 0.0458 (0.799) (0.768) Ln(asset) 0.297*** 0.291*** (0.0950) (0.0956) d_price1*Ln(land) 0.406* 0.443* (0.238) (0.246) d_price2*Ln(land) -0.368 -0.279 (0.556) (0.510) Ln(land) 0.151* 0.272*** (0.0852) (0.0988) Ln(land)^2 0.0542*** (0.0167) Average years of schooling 0.0147 0.0152 (0.0263) (0.0262) Household size 0.0765** 0.0686** (0.0339) (0.0343) Average age -0.00367 -0.00366 (0.0103) (0.0103) Constant 7.265*** 7.360*** (1.554) (1.554) Observations 762 762 Sample All All Provincial dummies included but not reported YES YES R-squared 0.247 0.253 *** p<0.01, ** p<0.05, * p<0.1 Standard errors are heteroscedasticity robust and clustered at the village level, and are reported in the parentheses. 32 Table 6: Impact of anticipated vs. unanticipated price shocks (first-stage regression on price expectation) (1) VARIABLES d_price d_price_t-1 -1.137*** (0.257) TR_t-1 -0.0106*** (0.00206) AR_t-1 0.642* (0.374) SD_t-1 -0.000631*** (0.000170) Share of rice in 07 0.249** (0.101) Share of maize in 07 0.777*** (0.247) Share of estate crop in 07 -0.492*** (0.0713) Share of horticulture crop in 07 -0.465*** (0.0970) Average age -0.00116 (0.000849) Average years of schooling 0.00275 (0.00196) Househouse size -0.00378* (0.00194) Yrs of schooling x Share of maize -0.0297 (0.0214) Yrs of schooling x Share of cassava -0.00923 (0.00955) Yrs of schooling x Share of estate crop -0.00956** (0.00381) Yrs of schooling x Share of horticulture crop 0.00224 (0.00444) Constant 0.632** (0.296) Observations 737 Provincial dummies included but not reported YES R-squared 0.843 *** p<0.01, ** p<0.05, * p<0.1 Standard errors are corrected by two-step bootstrap estimator with 2,000 replications, and are reported in the parentheses. Farmers with fixed price arrangement (Tebasan and Ijon) with traders are excluded from the sample. 33 Table 7: Impact of anticipated vs. unanticipated price shocks (second-stage regression on price expectation) (1) (2) (3) VARIABLES d_lninv d_lninv d_lninv Anticipated 7.513* 9.390** 10.01** (3.964) (3.890) (3.921) Anticipated * Ln(asset) -0.504** -0.566** -0.576** (0.231) (0.229) (0.228) Anticipated * Ln(land) 0.468 0.506* 0.496* (0.291) (0.282) (0.283) Anticipated ^2 -0.973** -1.258*** (0.461) (0.480) Unanticipated -3.507 0.417 -0.0391 (8.176) (8.255) (8.219) Unanticipated * Ln(asset) 0.0458 -0.124 -0.0866 (0.484) (0.489) (0.487) Unanticipated * Ln(land) -0.0443 0.207 0.203 (0.467) (0.466) (0.465) Unanticipated ^2 4.440 4.775* (2.758) (2.796) SD_prediction error -0.0906 (0.0702) Ln(asset) 0.283*** 0.297*** 0.304*** (0.0905) (0.0898) (0.0896) Ln(land) 0.298*** 0.289*** 0.293*** (0.0991) (0.0966) (0.0964) Land ^2 0.0636*** 0.0594*** 0.0603*** (0.0223) (0.0221) (0.0222) Average age -0.000654 -0.00205 -0.00236 (0.0110) (0.0110) (0.0110) Average years of schooling 0.0279 0.0192 0.0179 (0.0240) (0.0243) (0.0244) Househouse size 0.0655** 0.0725** 0.0708** (0.0318) (0.0318) (0.0322) Constant 7.325*** 7.064*** 6.891*** (1.527) (1.513) (1.522) Observations 737 737 737 Provincial dummies included but not reported YES YES YES *** p<0.01, ** p<0.05, * p<0.1 Standard errors are corrected by two-step bootstrap estimator with 2,000 replications, which are reported in the parentheses. Farmers with fixed price arrangement (Tebasan and Ijon) with traders are excluded from the sample. 34 Figure 1: Locations of surveyed villages 35 maize rice 1080 7080 1080 7080 20 04 m 20 1 2211.08 10211.08 2211.08 10211.08 05 20 20 m1 04 06 20 m1 20 m1 05 Graphs by prov 20 m1 Graphs by prov 07 Dynamics of rice price m 06 20 1 08 20 m1 Dynamics of maize price 07 20 m1 20 m1 09 08 Menado 20 m1 20 m1 10 09 m Banjarmasin 1 Menado 20 m1 10 m 1 Banjarmasin 20 04 20 m1 20 05 04 m 20 m1 20 1 06 05 20 m1 20 m1 07 06 20 m1 20 m1 08 07 m 20 m1 20 1 08 09 20 m1 20 m1 09 Lampung Semarang 10 m 20 m1 1 Lampung 10 Semarang m 36 1 20 04 mydate mydate 20 20 m1 04 05 20 m1 20 m1 05 06 20 m1 m 06 20 1 07 20 m1 20 m1 07 08 20 m1 20 m1 08 09 20 m1 m 09 20 1 Surabaya Makassar 10 20 m1 m Surabaya 10 Makassar 1 m 1 20 20 04 04 20 m1 20 m1 05 05 20 m1 20 m1 06 06 20 m1 20 m1 07 07 m 20 m1 20 1 08 08 20 m1 20 m1 09 09 20 m1 20 m1 Mataram Mataram 10 10 m m 1 1 estate cassava 500 5000 500 5000 20 04 m 20 1 116.74 1316.74 116.74 1316.74 20 05 04 20 m1 20 m1 05 06 20 m1 20 m1 Graphs by prov 06 07 Graphs by prov m 20 m1 20 1 07 08 20 m1 20 m1 Dynamics of cassava price 08 09 Menado 20 m1 20 m1 09 10 m Banjarmasin 1 Menado 20 m1 10 m 1 Banjarmasin 20 04 20 20 m1 04 05 20 m1 m 05 20 1 06 20 m1 06 20 m1 07 20 m1 07 20 m1 08 20 m1 m Dynamics of price for estate plantation crop 08 20 1 20 m1 09 09 20 m1 Lampung Semarang 20 m1 10 m Lampung 10 1 Semarang m 1 37 20 04 mydate 20 mydate 04 20 m1 20 m1 05 05 20 m1 20 m1 06 06 m 20 1 20 m1 07 07 20 m1 20 m1 08 08 20 m1 20 m1 09 09 m 20 m1 20 1 Surabaya Makassar Surabaya 10 Makassar 10 m m 1 1 20 20 04 04 20 m1 20 m1 05 05 20 m1 20 m1 06 06 20 m1 20 m1 07 07 20 m1 m 08 20 1 08 20 m1 09 20 m1 09 20 m1 Mataram 10 20 m1 Mataram m 10 1 m 1 Dynamics of price for horticulture crop Banjarmasin Lampung Makassar Mataram 914.185 314.185 20 m1 20 m1 20 m1 20 m1 20 m1 20 m1 1 m 04 05 06 07 08 09 10 20 hort Menado Semarang Surabaya 914.185 314.185 20 m1 20 m1 20 m1 20 m1 20 m1 20 m1 1 20 m1 20 m1 20 m1 20 m1 20 m1 20 m1 1 20 m1 20 m1 20 m1 20 m1 20 m1 20 m1 1 m m m 04 05 06 07 08 09 10 04 05 06 07 08 09 10 04 05 06 07 08 09 10 20 20 20 mydate Graphs by prov (Source) Bulog, BPS Figure 2: Prices of rice, maize, cassava, estate crop, and horticulture crop in Indonesia 38 8 6 4 2 0 -.5 0 .5 1 1.5 2 d_price1 d_price2 Figure 3: Distribution of food price variable (N=762) 39 .226594 Fraction 0 -4.243 1.61256 sd_error Figure 4: Distribution of 𝑆𝐷�𝑟𝑒𝑑𝑖�𝑡𝑖𝑜𝑛 𝑒𝑟𝑟𝑜𝑟 40