The county yields plugged into the Agricultural Risk Program (ARC) calculation have come under fire recently for perceived inaccuracy and varying dramatically across nearby counties. I was there when the last farm bill was written, there were concerns about the county yields, but a lot of people – Hill staffers, farm groups, and political appointees said surely the USDA can find a way to do this. Basically, this is a statistical problem and most of us hate statistics. I will try to shed some light here.
Be careful what you wish for
NASS uses a statistically survey approach to estimating yield. It is probably about as cost effective as one can approach the task. Pollsters, marketing firm, and researchers use these techniques all the time. But here is what you must know. NASS reported hundreds of county corn yields and then crop reporting districts, state, and national aggregates for 2015. Note also NASS did not report other counties due to small samples. The fundamentals of statistical surveying imply the accuracy of NASS estimates increases with each higher level of aggregation. The bottom line is while state and national numbers are highly credible in most cases, lower levels will simply be less accurate. Well-established rules of survey sampling dictate the primary way to get better NASS county yield estimates is to send more surveys into the county which will cost money and will necessitate greater respondent burden. This is an increasing problem over time as there are fewer farmers to be surveyed. So even if attempted it might not work.
RMA data is great – where there is a lot of it
What about RMA data? RMA does collect yields from participants and in many locations reaches 80 to 90 percent participation rates. But participation is not as randomized like the NASS survey, and in a county with an 80% participation rate one may ask what are the characteristics of the 20% not in the program. Are they the best yielding farms or the worst or neither? I am unaware of research that answers this question. I will note the RMA has develop their own county yield estimates for use in area insurance products including STAX and SCO. But they also encounter counties with limited crop insurance participation and thin data.
Statistical stews
Note that RMA does not mix NASS and their own data so that the historical benchmark is consistent with the covered year. I have looked at this some in the past and the RMA and NASS data often do not seem to match up. The RMA data was sometimes lower and sometimes higher than NASS yields. When NASS and RMA are mixed, you get a statistical stew that probably no one can sort out.
Farmers prefer individual protection if they can get it
Even if we get amazingly accurate county estimates will it be enough? I doubt it. It is pretty clear farmers want protection that is very highly correlated with their own yield – so the county triggers when the farm needs it. In 2014 before the introduction of STAX and SCO only two percent of crop insurance acres where insured with area insurance plans. Why? In part, there are real and perceived variation of yields within counties. A grad student in our department just defended a thesis showing a dramatic lack of correlation of farm yield with county yields in some counties. In one county, she found the farm-county yield correlation ranged from 0.18 to 0.93 (perfect would be 1.0). At 0.93 you have a pretty good risk management tools. At 0.18 you are pretty close to having payments with no relationship with farm losses. Remember ACRE with a state trigger was adopted in 2008. The fix was county yield in 2014. What next?
A jumpy clutch
My Dad started me on a Farmall Super C tractor. It had what he called a jumpy clutch, which meant it went from disengaged to engaged in what felt like a ¼ inch of release. Many fail to recognize that ARC goes from no payment to maximum payment with a 10% change in revenue. This mimic a design that my colleagues Barry Barnett and Steve Martin drew up for Steve’s dissertation many years ago. This differs from crop insurance the triggers at a given coverage and then reaches maximum payment at zero yield. The 10% range in ARC makes payments react quickly to slight differences in county yield. So county A has a revenue 14% below average and neighboring county B has a revenue 5% worse. County A gets zero ARC payment and County B gets half the maximum.
‘Fair Boundaries’
The average county in the United States is 997 square miles while the largest county in the lower 48 states is San Bernardino County California at 20,105 square miles. In Oregon the largest county is 23 times larger than the smallest county. All this just points out that counties in the U.S. are not defined in anything like equitable agricultural regions. This impacts the magnitude of payments and the correlation of farm-county yields. County size matters but so does crop acreage and heterogeneity within the area.
So what next?
Does USDA need to produce three slightly different county yields – the NASS, RMA, and FSA number? Compromises in the farm bill probably created some of this confusion.
Georeferenced data may help a lot someday. That day is nearing as USDA migrates to using more common land unit information for RMA and FSA. Layering of soil, crops and other information may give us the ability define areas based more sophisticated grouping. Here at Mississippi State we are working with National Commodity Crop Productivity Index (NCCPI) data that makes me hopeful.
But in the end, declining price guarantees in the ARC Olympic average for 2017 and beyond may make this a less important issue anyway. The Congressional Budget Office projects a dramatic decline in ARC payments for many crops for the 2017 year and beyond. This means less likely payments and smaller payment if they do occur.
Recent Comments