Incomplete Credit Markets and Commodity Marketing Behavior

1
Incomplete Credit Markets and Commodity Marketing
Behavior
Emma Stephens (Pitzer College) and Chris Barrett
(Cornell University) May 1, 2009
seminar Lafayette College
2
Motivation

• The Sell Low, Buy High Puzzle
• In developing countries, sharp seasonal grain
  price fluctuations are common.
• Few farmers take advantage of the resulting
  arbitrage opportunity.
• Indeed, many agricultural households do the
  opposite sell low and buy high.
• Why?

3
Hypothesis

• Hypothesis
• Puzzling commodity marketing patterns arise due
  to a displaced distortion from missing credit
  markets (Barrett 2007).
• Liquidity-constrained farmers use commodity
  markets as if they were a source of seasonal
  credit. The resulting terms of trade losses are
  akin to an interest rate on a seasonal loan.

4
Hypothesis

• Competing explanations
• Need to rule out alternative explanations
• Impatience /inferior returns to grain storage
• Unlikely. Bank deposit rate 5, seasonal maize
  price increase 44.
• High cost of storage and social taxation
• Unlikely. Inexpensive storage technologies
  widely used commonly limit losses to 1-2
  improving storage would yield far higher returns
  than sell low, buy high.
• Price risk
• But price risk aversion should generally lead to
  precautionary storage as hedging behavior.

5
Conceptual model

• Conceptual behavioral model
• Representative agricultural household both
  produces and consumes basic grain.
• Chooses consumption, production and storage so
  as to maximize intertemporal utility, defined
  over seasons, with
• (i) time-varying grain prices,
• (ii) transactions costs to market participation,
  (iii) possible liquidity constraints
• (plus the usual time and budget constraints).

6
Conceptual model

• Seasonal Household Market Participation
• In the absence of binding liquidity constraints
• Grain supply and demand functions in each season
  are a function of current and expected future
  prices.
• Transactions costs generate a price band and
  household-specific shadow price some households
  buy, others sell, others autarkic.
• Given exogenous seasonality in prices, demand
  increases (decreases) in the post-harvest
  (hungry) period due to low (high) prices.
• With constant transactions costs, if household
  participates in market, it should be canonical
  arbitrage buy low (post-harvest period) and/or
  sell high (lean period), smoothing consumption
  and maximizing intertemporal welfare.

7
Conceptual model

• Seasonal Household Market Participation
• With binding liquidity constraints post-harvest
• Post-harvest grain demand falls.
• Yet can no longer smooth consumption across
  seasons kinked Euler function means household
  optimally stocks out in post-harvest.
  Consumption no longer a function of expected
  future prices, just current. More likely to
  sell, even with low prices.
• Having stocked out post-harvest, have to buy in
  lean season to survive, in spite of higher
  prices.
• Result sell low, buy high behavior. When
  liquidity constraints bind, expected future
  prices and income no longer condition current
  choices. Arbitrage disrupted even though
  households recognize price seasonality.

8
Conceptual model

• Behavioral predictions
• Standard
• - Sales (purchases) increase (decrease) with
  market prices.
• Sales are increasing in productive assets (land,
  education)
• Novel
• - Greater income and credit access reduces
  (increases) the probability of harvest period
  grain sales (purchases) and of lean season
  purchases (sales).

9
Estimation strategy

• Challenges
• Transactions costs and shadow prices unobserved.
• Market participation behaviors correlated within
  and across seasons for a given household. Need
  systems estimator allowing for different
  parameters for season and market participation
  regime.
• Transaction volume decisions not independent of
  (discrete) household market participation
  choice. Potential for sample selection bias.
• Transaction volume censored.

10
Estimation strategy

• Approach
• Use Yens (2005 AJAE) multivariate sample
  selection model (MSSM), a switching estimator for
  censored demand systems
• - system of 4 (binary) market participation and
  4 (censored) marketed quantity equations one
  each for harvest/lean season purchase/sale

Marketed quantity Discrete market
participation Log(qs,ni) xsn,ißsn?sn,i if
zsn,iasnµsn,i gt 0 0 if
zsn,iasnµsn,i 0 where n season, n sale
or purchase, µ, ? MVN(0, ?) with ? the 8x8
covariance matrix
11
Estimation strategy

• Key independent variables
• Operationalize liquidity constraints using
• Credit access instrumented due to endogeneity
  using a probit identified using distance measures
  
• Off-farm cash income
• Identification
• We identify the discrete participation
  (selection) equations using fixed transactions
  costs

12
Data

• Survey data
• 2005 survey by Tegemeo Institute
• N1682 households in western Kenya
• Choice-based sample (corrections made)
• Monthly purchase and sale volumes and prices,
  July 2004-June 2005. We discretize this into two
  seasons harvest (July-January) and lean
  (February-June).
• Credit use and standard household data.

13
Data

• Maize is staple crop.
• Rudimentary, rainfed cultivation of, on average,
  only 2.3 acres.

Simple at-home storage, average value KSh859
(US12). Yet 87 report no maize storage losses
and mean loss for rest is lt8. 80 of households
had no stored maize at the time of survey.
14
Data

• Buy low/sell high is most common pattern (49.7)
  for those with seasonal net sales average loss
  29.3.
• Canonical intertemporal grain arbitrage rare (2
  )

15
Data

• Liquidity
• Binary indicator of credit obtained, whether for
  non-agricultural (typically consumption) purposes
  or for agricultural inputs. Access highly
  limited, only 28.5 get it.
• Credit use is a highly imperfect proxy, but the
  only one available in these data. Supplement
  with off-farm cash earnings (salary and
  self-employment).
• Instrument for credit access using distance
  measures to local markets and services. Simple
  first-stage probit to predict credit access.
  Standard results credit increases with income,
  education, longevity in village, etc., decreases
  with distance.

16
Results

• With respect to the liquidity variables of
  interest, effects generally consistent with our
  core hypothesis.
• Credit use associated with reduced likelihood of
  harvest season sales and lean season purchases
  and with increased likelihood of hungry season
  purchases.
• Credit use and off-farm income associated with
  increased purchase volumes (conditional on
  purchasing) in both seasons.
• Magnitude of credit/off-farm effects generally
  similar.
• Only odd result off-farm income associated with
  reduced likelihood of harvest season purchases.

17
Results
Corrected standard errors in parentheses.
18
Results

• With respect to conventional explanatory
  variables, effects are as one would predict
• Purchase volumes decreasing in price, sales
  volumes increasing in price.
• Probability of maize sales (purchases)
  increasing (decreasing) in land owned and sales
  volumes increasing in land holdings.
• Value of storage facilities has no statistically
  significant effects on marketing patterns.

19
Results

• Cross-equation correlations also intuitive
• Correlation of entry and purchase decisions for
  same season and marketing position analogous to
  an inverse Mills ratio are strongly positive.
  Households in the market transact more than a
  randomly selected household.
• Maize purchase volumes strongly negatively
  correlated with likelihood of sales in either
  season.
• Strongly positive interseasonal correlation in
  both sales and purchases volumes. No significant
  correlations between purchase and sales volumes.

20
Conclusions

• Yens MSSM works well in tackling this complex
  market participation behavior estimation problem.
• Liquidity constraints and seasonal
  quasi-borrowing seem the bestexplanation of the
  sell low, buy high puzzle. Inferior returns,
  high storage costs and price risk seem
  implausible causes.
• Financial market failures spill over into
  commodity markets by inducing stock-outs and
  breaking of standard intertemporal arbitrage
  conditions. Damages small farmers ability to
  accumulate agricultural profits and invest.

21
Thank you for your time, interest and support!
22
Harvest Season Market Participation
23
Lean Season Market Participation