NFAPP's Newsletter September 1996

Featured Articles :

"Forecasting Retail-Farm Margins for Fresh Tomatoes: Econometrics vs. Neural Networks", by Timothy Richards and Pieter van Ispelen
Topic: Retail-Farm Margins Tomatoes/Application Neural Networks

"Legislative Update.....", by Pamela Mischen
Topic: United's Washington Public Policy Conference

"Market Watch.....U.S. Apple Market Update", by Richard Adu-Asamoah
Topic: The U.S. Apple Fall Market

Existing econometric models of the retail-farm price spread are constructed with the objective of determining what factors are most important in causing farm and retail prices to differ. However, agribusiness managers and regulators alike need a tool that will enable them to forecast the behavior of margins in the future to determine the effects of price changes at one level on returns at the other. Current research shows that retail prices, expenditures at the retail level, marketing costs, past margin values, price variability, and concentration in the output or input markets all influence price spreads. Typically, however, this research makes simple assumptions about how these variables affect the margin, most assuming that margins change one-for-one with changes in retail prices, expenditures, etc. However, this is not likely to be the case when growers' risk attitudes, past margin values, or market structure issues are taken into account. Therefore, there is a need for a better forecasting to accommodate the true complexity of this problem.

Neural networks are particularly powerful forecasting tools when complex problems such as this arise. Using mathematical models that simulate the way the human brain operates, neural networks are able to "learn" the pattern of relationships between a set of input variables (retail price, risk, etc.) and an output variable (retail-farm margin). As such, they are often able to outperform traditional econometric forecasting methods. Neural networks are, however, often criticized for their inability to contribute to policy analysis or provide results that are as simple to interpret as existing approaches.

Advances in neural network software have, at least partially, cured this problem. By conducting sensitivity analysis with the input variables (risk, retail prices, etc.), their incremental effects on the output variable (margins) can be found numerically. Moreover, many software packages provide estimates of the "contributions" of input variables - or the amount of variation in the output variable that is attributable to changes in each input variable.

NFAPP's objective in conducting this research is to compare the ability of an econometric model and a neural network approach to forecast fresh tomato margins. Forecast performance is defined two ways - the ability to predict future margin values as accurately as possible, and to be able to predict when the margin series "changes direction" or swings up or down in one period. Just because one approach is able to forecast accurately does not necessarily mean that it will do well in predicting such turning points. As a secondary objective, NFAPP researchers use the features now available in neural network modeling software to do policy analyses similar to their "old style" models.

Data for this study consist of monthly margin observations from January 1980 to December 1994. For both the neural network and econometric models, training (estimation) uses the first twelve years of observations, while out of sample forecasts use the data from January 1992 to December 1994. All price and quantity data are from Tomato Statistics (USDA), while labor costs are from the Bureau of Labor Statistics, while the other data were taken from the WEFA Group database. Estimates of the econometric model are found using SHAZAM 7.0 and Ward Systems' Neuralshell II software is used to train and forecast the neural network model. The results from applying this procedure to the tomato data provided some surprises.

Comparisons of the forecast accuracy of the two methods, however, yielded no surprises. Using two measures of accuracy, the neural network consistently outperforms the econometric model. In fact, the value of each measure for the econometric model was about double that of the neural network. However, the econometric model is at least as good, and probably better, at predicting turning points in the margin series. This ability is key to the economic value of a forecasting tool. Often, money is made or lost in simply "guessing right" as to what direction the market will go, rather than knowing precisely what the price will be tomorrow when the price has been rising for days. Therefore, the true business value of neural networks may be significantly lower than once thought.

Second, the relative importance of the input, or explanatory variables differed between the two modeling approaches. For example, a time trend, tomato packing wages, and past values of the margin are the most important variables in explaining the margin according to the econometric model. However, these variables rank near the least important factors that the neural network uses in forecasting the margin.

Although the results from using each approach are considerably different, NFAPP sees the two tools as complementary rather than substitutes. Information from the neural network can be used to improve our econometric models and vice versa. For example, the econometric model assumes that changes in the margin at low wage rates are the same as changes at higher wages. However, the neural network shows that the relationship is more accurately described as U-shaped. We hope to use neural networks to analyze many more issues of interest to fruit and vegetable growers, packers, and retailers in the future.

United Fresh Fruit and Vegetable Growers' annual Washington Public Policy Conference gave members of the produce industry an opportunity to voice their policy concerns on the Hill. While the 104th Congress is winding down, United is gearing up for the upcoming legislative session. Of concern to the group are three major policy areas: 1) immigration reform, 2) the extension of methyl bromide use, and 3) continued funding for the Market Access Program.

Both the House and Senate passed versions of H.R. 2202, the Immigration Control and Financial Responsibility Act. House conferees were appointed on September 11 and a conference is expected the week of September 16. United has requested that the conference committee adopt: Senate language calling for a GAO study of the H-2A Program;
* Senate language regarding the standard of proof for employment documentation;
* House language to establish a 10 day grace period to correct I-9 Form mistakes;
* House language to assure protection from unreasonable employer sanctions;
* House language for pilot verification systems;
* Senate language requiring States that refer applicants to verify eligibility.

More importantly for fruit and vegetable growers, is legal immigration reform which was taken out of the current Bill. United is urging the creation of a workable guest worker program as one of its main goals for the 105th Congress.

As the Opening Speaker for United's Conference, Sen. Rick Santorum (R-PA) spoke of his wish to see government's role in agriculture reduced. Santorum supports opening markets as a strategy for agriculture rather than subsidies and, although he did vote for its passage, favors a reduction in the Market Access Program.

During a meeting with Majority House Agriculture Committee staffers, outlined the top priorities for the next legislative session.
* Revisiting the Research Title of the Farm Bill
* Crop Insurance for fruits and vegetable and the Noninsured Assistance Program (NAP)
* Impact of NAFTA on the fruit and vegetable industry
* Country of Origin labeling proposed by Rep. Bono (R-CA)
* Continued implementation of PACA
* Implementation of "flexibility" as a result of the Farm Bill and impact on fruit and vegetable acreage
* Generic Research and Promotion language from the Farm Bill

The Conference wrapped up Thursday with a point/Counterpoint Discussion between Sen. Byron Dorgan (D-ND) and Rep. Mark Foley (R-FL). Both agreed that there was a need to reform the estate tax which makes its difficult for subsequent generations to continue operating a family farm without incurring substantial debt to pay the estate tax. When asked about employment and the future of the American economy, Sen. Dorgan favors eliminating the tax break that is given to businesses that take their operations overseas resulting in fewer jobs for American workers. Rep. Foley preferred to concentrate on deficiencies in NAFTA which allow Mexico an unfair advantage in the tomato and citrus industries. Foley believes that we need to make the U.S. competitive first by dealing with differences in labor regulations before allowing free trade.

Larger Fall-season Crop in the Western States May Not Depress Grower Prices:

Supplies of fall-season apples are expected to increase in 1996 over 1995 (USDA-NASS). Midwestern and northwestern apple growers are expecting a larger apple crop following a short crop in 1995. Total U.S. utilized apple production (all seasons) for 1996, however, is estimated at 5,350,000 short tons (USDA-NASS), down by about 2% from 1995 production. Cold weather, excessive moisture and strong winds in some central and eastern states delayed or prevented adequate blooming, and have reduced crop potential in these regions (an estimated 15 and 29 percent crop reduction in the central and eastern states , respectively).

Larger fall-season crops are expected in Washington and California, but fruit size will be smaller in Washington. About a 15% larger crop in the western states is predicted. This larger crop, and the expected smaller fruit size in Washington state, may depress fresh-market apple prices in the short run. This may not, however, lead to lower grower prices all year as continued tight supplies of processing apples and apple juice stocks will keep prices strong. A strong domestic and export demand for fresh apples will also help keep domestic fresh apple prices relatively high. Shippers in California, on the other hand, speculate that their prices may decrease. This year's California crop will have smaller-than-normal fruit size for Royal Galas, and some Granny Smiths. California's Granny Smith harvest may also overlap with Washington's cold-storage Granny Smith season as well. California expects its Granny Smith and Fuji crops to far exceed 1995 production levels (4.5 and 2 million cartons respectively, compared to 3.5 and 1.7 million carton last year), while the Gala crop may be slightly lower (1 million cartons, compared to 1.2 million cartons in 1995). August 1996 prices will be substantially lower than 1994, but grower prices will be, at least, as high as 1995 levels for August and September. Prices may fall after September, but will likely stay stronger than 1995 levels for the rest of the year.

Higher grower prices registered during January through July 1996, were the result of a smaller Fall 1995 crop, and an improved domestic and export demand. Average grower price for all apples in 1996/97 is estimated at $0.17 per pound. This price reflects the estimated lower U.S. 1996 crop, and continued tight supplies of processing apples.

Grower-Retail Margins Likely to Widen Through September:

A strong export demand will help maintain reasonably high domestic grower and retail prices. The gap between retail price and grower price, however, will likely continue to widen through September. The 1996/97 average grower price of $0.17/pound will be possible only if 1995 production levels in importing countries, and early 1996 (January through July) export demand trends are maintained. Europe and Mexico have not yet fully recovered from the bad weather in 1995 that substantially reduced crop size. Through 1997 the strength of export demand, in addition to improved domestic demand, will contribute substantially to strong grower prices. Efforts by major producing states to improve market share overseas will continue to pay off. The California Apple Commission is using Market Access Program (MAP) funds to promote California apples in Taiwan. In addition, California is teaming up with Idaho, Michigan, New York, Pennsylvania and Virginia to promote apples in Brazil. The Washington Apple Commission is also continuing its promotion efforts abroad, and by August 10 Washington had already started exporting from the 1996 crop.

To date, U.S. apples are exported to almost all corners of the globe. Established and growing markets include Canada, Columbia, Costa Rica, El Salvador, Egypt, Guatemala, Honduras, Hong Kong, Indonesia, Iceland, Kuwait, Malaysia, Mexico, Philippines, Russia, Singapore, Taiwan, Thailand, and United Kingdom. China and Japan are still potential markets for U.S. apples.

Relatively stable retail prices are expected nationwide for fresh apples, in 1996, following the larger crops in the western states, and anticipated increase in imports. The gap between retail price (Red Delicious variety) and grower price of fresh-market apples will likely continue to widen through September. Retail prices will increase as a result of increasing transportation costs, refrigeration and storage costs, cost of other marketing services, and a strong demand. The expected strong domestic fresh-market retail prices will, however, contribute only marginally to any gains in grower-level prices.

Back to NFAPP's Newsletter Articles Page

Back to NFAPP's Homepage