Government production estimates.
No one likes them.
Everyone disagrees (or at least most of the time).
And I know plenty of people who just don’t trust the numbers…
But eventually, we just admit that’s the way it is… then it’s time to move on.
It usually takes about two months before the market stomachs the numbers.
But the revisions are what kills us.
Take the numbers from Statistics Canada for example.
Chuck Penner from LeftField Commodity Research tweeted a chart about StatsCan’s track record on estimating and then revising canola production in Canada. 
Honestly, for me, the numbers are shocking.
I watch the markets closely, but I’ll admit that revisions tend to get baked into the actual market pricing structure and so I don’t follow the revisions carefully enough.
And I think a lot of other people weren’t either.
That’s why I spent a few hours digging through these revisions.
I looked at all the actual reports of production estimates of principal crops from July 2012 to July 2017.
So, not including the 2017 growing season (because it’s still ongoing), I reviewed five years of reports.
In that period, the agency has released 22 reports, not including the satellite data series.
In all of those reports, the revisions are constant. Not just quarter-to-quarter, but the changes come up to 2 years after the fact.
I guess you can give some benefit of the doubt here, but it’s the size of the revisions that continues to shock me.
As you’ll see, very rarely does StatsCan revise crops smaller.
Using the past five years as the benchmark, grain, oilseed, and pulse crop production almost always increase in Canada.
And not by just a little bit either.
What the Canola….
Take canola for example.
In the last four years, StatsCan has underestimated the Canadian canola crop by an average of 3.2 million tonnes.
This trend means that in the past four years the Canadian canola has grown by an average of 24% from StatsCan’s first estimate.
A simple analogy to help you understand this sort of difference, it’s like climbing a 10,000-bushel bin, looking in and seeing that’s full. Then, after emptying said bin, you would look back in and see that one-fourth of the bin is still full.
2012 was the only year that StatsCan overestimated the Canadian canola crop. The July 2012 estimate suggested that 15.4 million tonnes of canola would be produced.
However, the number came in 10% smaller at 13.8 million tonnes.
This revision has obvious implications on balance sheets (i.e., what’s being carried over into the next crop year).
This margin of error can suggest a lack of price impact.
For 2017, using the average margin of error for the last five years of +17.3%, the 2017 Canadian canola crop should eventually wind up at 21.35 million tonnes.
Just to add some perspective to that, StatsCan first estimated the 2016 canola crop at 17 million tonnes. The satellite/data estimate a month later called for an 18.3 million tonne crop. The final number was just revised to 19.6 million tonnes.
This year’s first estimate was 18.2 million tonnes. The recent satellite estimate was for 19.7 million tonnes of canola.
Pulses and Wheat Are Always Underestimated
In 2014, 2015, and 2016, StatsCan underestimated peas and lentils by an average of 11% and 17% respectively.
The Canadian peas production number usually climbs by an average of 355,000 MT from the first estimate in July to the final number 2 years later.
For lentils, the crop tends to have 263,000 extra tonnes added to its final production number. That’s another 5 or 6 Panamax boats going over to India.
To be fair, chickpeas usually are overestimated, and the crop comes in smaller than originally suggested. 15% lower to be exact. However, because the Canadian chickpeas crop isn’t huge to begin with, that 15% equates to a final number showing just 18,000 MT taken off from the first estimate.
For wheat, StatsCan certainly misses the boat on a spring-seeded product. Durum wheat production has been underestimated each of the past five years and by an average of 15%.
This equates, on average, to adding 777,000 MT to balance sheet from the first estimate.
For spring wheat, StatsCan has revised the crop higher in each of the past five years except for 2012. And even then, the adjustment was only a 1% decrease from the first estimate (or 212,600 MT).
Comparably, in the past four years, StatsCan usually increases the Canadian spring wheat crop by almost 1.75 million tonnes. This means that, on average, the Canadian spring wheat crop has grown by 8.5%.
The outlier was obviously 2013 when the spring wheat crop was increased by 5.4 million tonnes or 25% from the first estimate.
With that year removed from the analysis, the Canadian spring wheat crop has grown by an average of more than 6%, or 1.17 million tonnes more than the initial estimate.
Ideas to Improve StatsCan Crop Production Estimates
You might disagree with what you’ve read thus far and want to challenge me as to how StatsCan could be better.
Here are three ideas to get started…
1. StatsCan could take a page out of the book of their comparable, the United States Department of Agriculture
The USDA not only does phone call surveys but also adds in numbers from their actual fields in different regions across America. Further, the USDA incorporates satellite data into all of their estimates.
Also, they might want to consider updating their Data Sources and Methodology section.
As per that section, the questionnaire sent to farmers has evolved/been updated, but it was originally designed in 1908.  If the entire data-reporting process was available online, I wasn’t able to find it.
Conversely, the USDA’s NASS yield forecasting program has a 108-page document explaining their calculations. 
One. Hundred. And. Eight.
I scrolled and read through StatsCan’s Data Sources and Methodology in less than 10 minutes.
The only thing I don’t want the StatsCan to mirror from the USDA is their software tools.
Based on the budgets and bureaucracy, USDA is still probably using Microsoft Office 2007.
Also, we’ve heard stories that a prominent USDA wheat economist who plays a key role in monthly price reports still does his calculations with pencil and paper because he doesn’t know how to use a computer.
Statistics Canada should not do this.
2. Stop comparing data year-to-year.
Every farmer knows that each farming year is different from one to the next.
Does StatsCan know this?
Changes in soil moisture conditions, heat units, and more play a significant role in determining final yield and production numbers.
If anything, the statisticians should consider comparing the data against years with similar growing conditions.
If your baseline for data comparisons is a drought, and you have a wet year the next, how does this affect their models?
These are questions that many individuals have asked, but there isn’t much clarity into the way that they arrive to their decisions.
Methodology is very important, but it appears that for all of the data that Stats Can has access to on a daily basis, they don’t have a very good message in how similar years in the past compare. It seems as though their models do not or will not account for all of these variables.
Perhaps they just like to provide the baseline – you know – the easiest estimate. Or maybe they’re just as bureaucratic as the USDA and it takes them forever to adjust numbers because the analysts need permission from 14 people.
3. Simplify the survey by putting it online (in addition to phonecalls).
The average StatsCan survey with farmers for the field crop reporting series takes 18 minutes.
One must use codes to differentiate bushels from kilograms from metric tonnes and so forth. This is not easy to deal with.
I’m happy to build them a simple excel file that would automatically calculate the values of each unit of measurement as soon as they put in just one.
In fact, we’ve already done it for them. From a front-end perspective, you see it as GrainUnitConverter.com. However, the backend that powers the frontend is the table I just described above.
Also, having to refer to said codes flipping between pages is absurd. Someone doing the interview likely has to scroll up and done, which intuitively increases the human margin for error.
I think this survey should be put online. Simple drop-down menus would erase the need for multiple pages and codes.
Further, have a top-notch user experience/user interface designer take a pass on this. There are plenty in Ottawa.
I know because we’ve got two great ones working for FarmLead out of our Ottawa office.
Plus, the current Canadian federal government doesn’t seem to understand the term “budget” so they should easily be able to find some strong candidates. I’m happy to hire out the FarmLead UX/UI team to StatsCan to fix this survey for $1,000 / hour.
All sarcasm aside, I’m seriously offering to help to fix this survey but my time and my team’s time is worth something.
One Benefit to StatsCan’s Poor Estimates
One obvious takeaway here is that because of StatsCan constantly revising Canadian crops to make them larger, the market doesn’t price this in right away.
Thus, those bigger crops, in a sense, help farmers because the crop looks smaller than it really is. As such, grain prices should be higher than they’re really supposed to be.
Theoretically, this could be because farmers underestimate their own yields when asked about their production estimates. I’m sure there’s a lot of farmers who would admit to giving an estimate that’s on the very low-end range of possibilities.
Regardless of what you’re telling StatsCan, and, comparably, what StatsCan is telling everyone else, you still must sell or market the grain that you produce.
While the market price is derived by supply and demand factors, said supply factor is often the unknown of the two. Especially in the middle of a growing season.
This is part of the reason that I started FarmLead: help a farmer find the best possible price for their grain, regardless of the market conditions.
While StatsCan certainly has some work to do, let’s get to work on your grain marketing plan and post your next 10 % (or so) on the FarmLead Marketplace. We’re currently recommending the sale of fall cereals, feed grains, yellow peas, and milling oats.
On the horizon, there appears to be upside opportunities in other oilseeds and higher quality wheats. For the latter, it’s a given that you must know important quality parameters like protein, moisture, HVK, Falling Number, and vomitoxin or DON levels.
Get your wheat or any other grain tested by any of FarmLead’s independent partner labs on GrainTests.com today.
StatsCan isn’t doing giving out favours. Why not get your grain tested?