Monday, August 23, 2010

Week 21

The changes this week are modest this week and show modest recovery from the declines of last week. The uncorrected model was basically flat and the models are largely converging now that we are approaching September. The same behavior can be seen in soybeans where the two models are converging. While the corrected model improves while the uncorrected model declines we see that while conditions are deteriorated the last several weeks, it is less than that seen in a 'normal' year.
Cotton yields and production stabilized after several weeks of decline. The crop still looks like a large one and cotton prices have been strong as world textile demand has risen and crop production issues have occurred outside the US. So for cotton producers it looks like it could be a good year.

Tuesday, August 17, 2010

Week 20

Week 20 spreadsheet

Corn yields and production both fell sharply after weeks of steadily increasing values. Alternatively, soybeans were fairly stable with the two models continuing to converge as we approach late September.

Cotton also showed a significant decline in yield and production, which is a continuation of a decline in values over the last several weeks.

Thursday, August 12, 2010

Spreadsheet with state by state comparisons wth USDA

Tables with USDA comparisons are updated in the spreadsheet
Week 19 spreadsheet

Comparison to USDA

Not bad overall. I think we can start with a discussion of ' a big crop gets bigger'. This isn't just a feeling you get but I suspect a statistical fact. I've mentioned several times that the uncorrected model is if we finish the year in the current conditions, but because they more often decline than they improve or stay the same, the corrected model is the best guess for end yields. But what happens if conditions don't decline through the year? Well the corrected model would be making a downward adjustment for crop problems that never occur, and as we get closer and closer to the end of the season, the adjustment gets smaller and it would increase to meet the uncorrected model.

Now, given the way the USDA seems to do their estimates, in the parameters, there would be embedded this normal decline. A simple fact of estimation. Then in September, even if those input variables didn't really change, yields are likely to be slightly higher because the time left for problems to arise has declined. So in years where conditions stay strong, the crop does get 'bigger' as time goes by. No real surprise here.

I'm on the low side of corn yields but closer on production. I've put up the spreadsheet with state by state comparisons. There are some places where if I updated the model I would go in and look at some of the parameters. For instance, I'm only abandoning 188,000 acres in Minnesota. This is actually mostly silage and silage averages about 450,000 acres, so the USDA harvesting rate looks much more plausible. Also Nebraska, the model did poorly with yields last year and it appears to be doing the same this year. Conditions are similar to last year and last year produced 178 bu/ac. So I think I'm under predicting in Nebraska for production. In the end my off-setting errors in corn put me pretty close to the right national numbers, closer than most of the trade on production. But if you take apart the state by state numbers you could certainly find some disagreement in what I have.

For soybeans, I'd say I'm pretty close. The same issue of a big crop gets bigger apply here as well, and in cotton for that matter too. The USDA yield was on the upper end of the trade estimates as were my model results. I don't have the same abandonment issues in soybeans. Kansas (low) and Minnesota (high) seem to have offsetting errors, but certainly stand out as having significantly different yield numbers than USDA.

For cotton, USDA trimmed yields a bit but production rose slightly. The USDA number hit in the range of trade. Now my yield is above the USDA's but I also have a slightly higher abandonment rate than USDA. USDA shows about 300,000 more acres of cotton being harvested on the same planted area than I do. Almost all of this coming from Texas. I actually have higher yields than USDA in Texas but also higher levels of abandonment which makes the two models 150,000 to 250,000 bales lower than USDA for cotton production in Texas. It is stating the obvious, but this is the difference that matters and will drive the final numbers.

Tuesday, August 10, 2010

Vs. The Trade

I've obtained some updated trade estimates, so I thought I would update the graphics before the Thursday release by the USDA.

I'm in the center of the trade for corn for yield and production. I'm in the top half for soybeans, both yield and production and this time the trade appears to have jumped to the other side of me in cotton. Remember, the uncorrected model is if we finish the year with current conditions while the corrected model takes into account that conditions more often decline than they do improve. In really good years, however, conditions don't decline. The blue line (corrected model) would continue to rise if conditions are simply stable. Are we in one of those years? It is hard to say.

Monday, August 9, 2010

Week 19

I'm center of the range for corn, both in yields and in production. I must have a slightly higher harvest rate given my production numbers are slightly higher relative to the trade than my yields. For soybeans, again I'm in the range of trade, but in the top half.  For cotton, I don't have much trade information to compare myself to. I was well above the trade and yet spot on last month, but cotton is always tricky. I should note that in some weeks I don't have updates for New Mexico cotton conditions until after I post this, so I use last weeks conditions carried forward. This is small and doesn't effect the total much, but in case somebody out there is following New Mexico (and if so, you should know the average error for this state is probably not small).

For corn and soybeans I'm well withing the trade range so unless there is a big surprise on Thursday, I hope to be close. and as far as I can tell, I'll be closer than 1/2 the folks in 'the trade'. Note on cotton that my yield is a bit higher than USDA but I have a slightly higher abandonment rate. This abandonment rate is calculated based on condition, but yields and abandonment rates don't always move in opposite directions. If the top end of conditions stay the same but at the bottom end, some of the crop moves from poor to very poor, this may increase the abandonment rate as this area does not get harvested, but the remaining harvest crop is then on average in better condition and yields might actually rise. Given these relationships in cotton I tend to focus on the production number more. Abandonment is not nearly as volatile in corn and soybeans so the same problem does not arise here.

If you read back over the archive, the whole purpose is to construct a parsimonious model to predict yields. Something that doesn't take a lot of time and draws out the value of USDA-NASS statistics. I'll do a graphical update on Wednesday if I get new trade estimates and after the report comes out I'll do a comparison table filling in state by state USDA-NASS estimates and comparing them to my own.

Week 19 spreadsheet

Tuesday, August 3, 2010

Nebraska, a state to watch

I think Nebraska is a state to watch in these estimates. Last year was 178 bu/ac and conditions are better now than they were last year when yields were 178.

Last year 2-5-14-52-27
This year 1-4-11-57-27

I'm only getting 170 bu/ac out of the model. Was last year an unusual year? Have yields in Nebraska accelerated? Are the parameters I estimate for Nebraska showing to little difference between poor to good? (they are within 4 bushels of each other while excellent is ~40 bu/ac higher than that). I think this is one place where I could be badly off. Hopefully I've got other offsetting errors but if they did 178 bu/ac last year, can 180 bu/ac be out of the question this year?

Let me know what you think, post anonymously or send an email. I don't adjust the model in this simple exercise because I don't spend enough time analyzing all the information that is out there. These estimates were an attempt at being parsimonious, and this may be just one instance were some additional examination would lead one to a different conclusion about corn production in Nebraska.

More on state by state numbers, now for Soybeans

The 5 state numbers for soybeans show a similar pattern to corn below. Again Iowa sows the biggest changes so far. Again, I'm not ready to attribute this to greater fluctuations in soybean conditions just yet, it could also be the sensitivity of the parameters. Many analysts who use this data simply create a fixed weight index where Index=1*very poor + 2* poor + 3*fair + 4*good + 5*excellent and do this at a national level. I do each state individually and estimate the weights individually with the only constraint in estimation that they have to be non-decreasing with condition.

I also include a variable on trend and may include a variable on planting progress. It is the parameter on planting progress which got me into so much trouble last year as Illinois had the biggest planting delay in 4 decades but had reasonably cool and moist temperatures the rest of the year which lead to better yields than my simple model would have anticipated. The planting progress is basically an intercept shift. If you want to read more on Illinois crop yields from last year I'd suggest the folks at U of Illinois who do far more detailed models than I do here.

Monday, August 2, 2010

Short note on state by state numbers.

The Iowa yield numbers are significantly higher than those for Illinois but have also show significant variation so far. I don't attribute that to big changes in Iowa as much as I might attribute it to problems with equation parameters. Maybe there is a to big a yield difference between some of the condition categories?

If we lay out the Iowa equation, the Very poor category is the base.
when we go to poor we add +0, so no change in yields, I can't determine any real difference between very poor and poor. The same is true with fair at +0, but good adds +30bu per acre and excellent adds +66 bushels per acre. It could also be there there are just isn't a lot of Iowa that is in very poor or poor condition.

For Illinois we have (again with very poor as the base) poor +0, fair +26, good + 69.1, excellent 69.8 so a much different gradient depending on condition shifts.

Also to add to the discussion , here are the trend yields in corn and soybeans for the top 5 production corn states (in 2007) and also my home state of Missouri. These are the underlying annual trend growth I estimate in the equations. This should be adjusted for conditions and thus starting and ending year are less relevant, but probably not inconsequential.

State-corn trend yield-soybean trend yield

IL - 2.94 bu/ac - 0.47 bu/ac
IA - 2.66 bu/ac - 0.53 bu/ac
IN - 2.32 bu/ac - 0.63 bu/ac
NE - 2.05 bu/ac - 0.74 bu/ac
MN - 2.47 bu/ac - 0.24 bu/ac
MO - 2.21 bu/ac - 0.51 bu/ac

The numbers I show for Iowa would be a little lower than estimated last year and the ones for Illinois might be a bit higher, you can see what I used last year in the archives.

You can click on the table below to enlarge it. It is a state by state reporting of the corrected model production and yield for corn as of this week. I'll update it in next week. Feel free to print it out and pencil in your own guess, how about some anonymous postings on what YOU think the USDA yield will be?

Week 18

Corn and soybeans estimates are converging as we would expect as we approach the end of September where the equations were estimated. So the corrected model shows a modest increase for corn yields while the uncorrected model shows a modest decline. Why? Because although conditions in aggregate may have slipped slightly this week, the decline was less than the historical average during this week and thus the corrected model, which is the best guess, shows a bit of improvement.

Just to clarify, if conditions NEVER changed through the year, the uncorrected line would be a flat line while the corrected model would rise to meet it.

The hot weather in cotton country shows up with a decline in both cotton yields and production. I always hesitate to take one week as a indicator of actual change, so this one will bear watching. If unfavorable temperatures continue, we should see a steady decline here. I'll be posting again on Monday of next week and then again after the USDA report on the 12th to see how close I am to USDA