Pricing in the Coffee Industry: An Econometric Model of Prices to Growers
University of Wyoming
Professor David Aadland
April 26, 2018
Alex Belser
Introduction: Not only do the farmers at origin play an integral part in your morning cup of
coffee but they also depend on coffee as a means to provide for their families. The global price
of coffee is characterized by extreme volatility—leading one to question if farmers’ lives are
inherently tied to this unreliable market. This research paper seeks to answer this using
econometrics: does the global commodity price of coffee impact prices paid to growers? An
econometric model is developed for the use of hypothesis testing to address this question.
Literature Review: There is a large existing literature which examines the global coffee market
and the role of commodity markets in pricing. While examining the role of institutional
arrangements in transmitting prices to coffee producers, Cárdenas (1994) finds that market price
volatility affects producers at varying degrees dependent on domestic government intervention.
This research provides a launching point for investigating the relationship between global coffee
prices and the prices paid to growers, as it illustrates the interconnectedness of the two.
In their 2007 book, Benoit Daviron and Stefano Ponte illustrate the paradox of “the
coffee boom in consuming countries and the coffee crisis in producing countries,” in which
producers find themselves trapped in the ‘commodity problem’ so long as they do not control
any of the intangible aspects of the experience that retailers possess. The authors break down the
global coffee market and provide insight into the value chain, suggesting that commodity market
prices have a significant impact on the farmgate prices growers receive.
Model: The dataset used for this model provides data for fifteen countries from the years of 1990
to 2017. The dependent variable in this model is the price paid to growers at the farmgate level,
GrowerPrices. The independent variables in this model are the global price of coffee,
WorldPrice, the average world retail price, RetailPrice, and the United States producer price
index for coffee, PPI. Definitions of these variables can be found in Table 1 below.
Table 1
Variable definitions and descriptive statistics
Variable Definition Mean Min Max Std Dev
GrowerPrices Annual average price paid to grower at farmgate level* 83.7 17.6 239.7 40.7
WorldPrice Annual average for global price of coffee (arabica) ** 127.8 60.4 273.2 52.2
RetailPrice Annual world avg. consumer price paid at commercial outlets † 5.3 3.7 7.0 1.0
PPI Annual average for U.S. producer price index by coffee †† 155.1 100.5 223.9 36.7
Notes: *indicates data from the International Coffee Organization (ICO) and are in (US cents/lb),
** indicates data from the International Monetary Fund, retrieved from FRED and are in (USD/ton),
† indicates data from the International Coffee Organization (ICO) and are in (USD/lb),
†† indicates data from the U.S. BLS, retrieved from FRED and represent an index where 1982=100
The empirical model takes the following form:
𝐺𝑟𝑜𝑤𝑒𝑟𝑃𝑟𝑖𝑐𝑒𝑠*+ = 𝛼* + 𝛽1𝑊𝑜𝑟𝑙𝑑𝑃𝑟𝑖𝑐𝑒+ + 𝛽5𝑅𝑒𝑡𝑎𝑖𝑙𝑃𝑟𝑖𝑐𝑒+ + 𝛽9𝑃𝑃𝐼+ + 𝑢*+
This paper intends to determine if farmgate prices paid to growers are dependent upon the
global commodity price of coffee. As such, the main hypothesis this paper seeks to test is that the
global price for coffee does have a significant impact on the prices paid to growers at farmgate
level. The hypothesis test will be set up with the null stating that WorldPrice does not have a
significant effect on GrowerPrice. A significance level of 5% will be used for testing this
hypothesis. The null and alternative hypotheses can be written as:
𝐻=: 𝛽?1 = 0
𝐻A: 𝛽?1 ≠ 0
In addition to the main independent variable of the global market price, WorldPrice, the
developed model also includes RetailPrice and PPI. The expected signs of the coefficients are as
follows. One would expect 𝛽?1 to be positive—as the global market price of coffee increases, the
prices growers receive should increase. 𝛽?5 is expected to be positive—as the average world retail
price of coffee increases, the prices growers receive should increase. The expected sign of 𝛽?9 is
not as clear, though one would expect it to be positive. This would be because as the prices final
coffee goods producers in the U.S. receive increase, the prices growers receive increase—
perhaps an indication of increased total value in the supply chain.
Potential econometric issues with the specified model are addressed as followed. This
model employs a linear functional form. A calculation of a pairwise correlation matrix indicates
there may be some multicollinearity between WorldPrice, RetailPrice, and PPI (refer to
Appendix 3). Graphs of residuals for individual cross sections may show potential
autocorrelation, though it does appear to be severe (refer Appendix 4). In case of any
inconsistencies in the variance of errors and thus heteroscedasticity, this model has been adjusted
to provide robust standard errors, clustered by countries. A Ramsey RESET test was conducted
to test for any indication of model specification error (see Appendix 5 for details). The obtained F
statistic of 0.1035 fails to reject the null hypothesis that the model has no omitted variables.
There may be some concern over endogeneity in this model as GrowerPrices may have some
effect on independent variables, such as WorldPrice or RetailPrice. It is important to
acknowledge these potential econometric issues in the model as the results are discussed.
Results: In this section, the effects of independent variables on GrowerPrices are examined, and
then the results from the main hypothesis test are discussed. A Fixed Effects OLS model with
robust standard errors was run to test the main hypothesis—refer to Table 2 below for the results.
Table 2
Results of FE Regression, clustered by group, number of groups=15, N=336
Variable Coefficient (Std Err) t-stat p-value
WorldPrice .6113516*** (.0741229) 8.25 <0.000
RetailPrice 6.146251*** (1.892088) 3.25 0.006
PPI -.0362522 (.1378161) -0.26 0.796
F statistic for country FE 72.34
Notes: *p<0.1, **p<0.05, ***p<0.01 Standard errors are in parentheses. Estimates for country fixed-effects are not
shown. The F statistic refers to the joint significance of the country fixed effects.
A t-test was calculated for the main hypothesis (refer to Appendix 6 for details) and the
calculated t-statistic (8.248) was larger than the critical t-value (~1.97). Subsequently, this result
rejects the null hypothesis that the global price of coffee, WorldPrice, does not impact prices to
growers, GrowerPrice, in favor of the alternative hypothesis. This is in line with our primary
hypothesis and our data suggest that the global price of coffee has a significant effect on the
prices growers receive at farmgate level. Refer to Appendix 7 for the conditional plot of
WorldPrice on GrowerPrices, which indicates a positive trend between the two.
The results of our regression show that the signs of the coefficients for WorldPrice and
RetailPrice are positive, which is consistent with our initial expectations. The coefficient for PPI
is negative, which goes against our initial expectation; however, this term is not statistically
significant at the selected 5% level. The significant coefficient estimates can be interpreted as
follows. As the average world retail price increases by one-cent per pound, the prices growers
receive can be expected to increase by roughly six-cents per pound, all else constant. As the
global price of coffee increases, the prices that growers receive can be expected to increase, all
else constant. Interpreting the magnitude of the WorldPrice coefficient estimate requires some
additional steps as the data is reported in US Dollars per ton. After conversion, the coefficient
estimate can be interpreted as such: a five-cent increase in the world price can be expected to
lead to a three-cent increase in the prices growers receive, all else constant. While these
estimates are shown to be statistically significant, they also appear to have economic, or practical
significance. Increases in the global coffee price will lead to a less than one for one increase in
the prices growers receive; however, those increases would still have a considerable impact on
the livelihood of farmers. Increases in average retail prices have a surprisingly large practical
impact on the prices growers receive—an increase of one-cent would lead to a five fold increase
in the prices farmers receive.
Conclusion: The results of regression model developed in this paper are consistent with the
initial hypothesis, suggesting that the global market price of coffee has a significant impact on
the prices that growers receive. The model also suggests that average world retail prices have a
significant effect on the prices that growers receive. The practical implications of the model
suggest that increases in retail price could be far more impactful in increasing farmgate prices
than increases in global price of coffee. While these results make intuitive sense, they should be
concerning to those of us who care about equality and economic development. This paper’s
findings suggest that the livelihoods of coffee farmers—primarily in the southern hemisphere—
are a plaything of the global commodity market for coffee’s incredible volatility, while
surrounding literature suggests that the roasters and cafés—primarily in the northern
hemisphere—are more immune to this volatility. As such, this research could be useful for
scholars hoping to explain pricing dynamics in the coffee sector as well as those in the specialty
coffee industry hoping to improve conditions for farmers. Future research might attempt to hone
in on these effects using monthly data rather than annual data, as this could have been a
limitation of this paper’s model.
Appendices:
Appendix 1: Plot of annual average global price of coffee (arabica)
Appendix 2: Plot of annual average farmgate prices to growers
Appendix 3:
Table 3
Pairwise correlation coefficients for individual cross section (Brazil)
Variable year WorldPrice RetailPrice PPI
year 1.0000
WorldPrice 0.5620 1.0000
RetailPrice .6487 0.8546 1.0000
PPI .8909 0.8448 0.8800 1.0000
Appendix 4: Residual graphs for individual cross sections
Appendix 5: Results of Ramsey RESET test
𝐻=: Model has no omitted variables
F(3, 329) = 2.07
Prob > F = 0.1035
Appendix 6:
Hypothesis test for WorldPrice, 𝛽?1
𝐻=: 𝛽?1 = 0
𝐻A: 𝛽?1 ≠ 0
D
The t statistic is 𝑡 = CE = .LMM5NMLI = 8.248 FG(CE) .=O9M11P
With 𝑛 − 𝑘 = 336 − 4 = 332 degrees of freedom and assuming a 2-tailed test, the critical t
value for a 5% significance level is 1.96 ≤ 𝑡\]*+ ≤ 1.98 and we reject the null hypothesis in
favor of the alternative.
Appendix 7: Conditional plot of WorldPrice on GrowerPrice
Works Cited:
Cárdenas, M. (1994). Stabilization and redistribution of coffee revenues: A political economy
model of commodity marketing boards. Journal of Development Economics, 44(2), 351-
380
Daviron, B., & Ponte, S. (2007). The coffee paradox: Global markets, commodity trade and the
elusive promise of development. London: Zed Books.
Gujarati, Damodar N., and Dawn C. Porter. Basic Econometrics. McGraw-Hill Irwin, 2009.
International Coffee Organization, Historical Data on the Global Coffee Trade—1990-2017,
retrieved from http://www.ico.org/new_historical.asp
International Monetary Fund, Global price of Coffee, Other Mild Arabica [PCOFFOTMUSDM],
retrieved from FRED, Federal Reserve Bank of St. Louis;
https://fred.stlouisfed.org/series/PCOFFOTMUSDM
U.S. Bureau of Labor Statistics, Producer Price Index by Commodity for Processed Foods and
Feeds: Coffee (Whole Bean, Ground, and Instant) [WPU026301], retrieved from FRED,
Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/WPU026301