30 Years of Guessing: Why Cotton Forecasts Have not Improved

Forecasts of world cotton supply, use, stocks and prices are no better today than they were 30 years ago, despite huge advances in information technology, communications and travel. Why hasn’t improved access to information translated into better insight?

First, the facts: one year-ahead forecasts of world cotton supply and less accurate today than they were in the 1980s. The absolute value of the percentage differences between ICAC forecasts of world production made each January and the eventual “correct” total averaged 3% from 1987/88 to 1991/92 compared with 7% for the most recent five seasons. For world cotton consumption, ICAC forecast errors for the season ahead made each January averaged 3% during 1987/88 to 1991/92 compared with 7% for the most recent five seasons .

The absolute value of the percentage differences between USDA forecasts of world production made each May and the eventual “correct” total averaged 2% from 1987/88 to 1991/92 compared with 3% for the most recent five seasons. For world cotton consumption, USDA forecast errors each May for the season ahead averaged 2% during 1987/88 to 1991/92 compared with 7% for the most recent five seasons .

ICAC Track Record

USDA Track Record

Townsend 30 Years of Guessing

The private sector is doing no better. Within-season volatility of the Cotlook A Index has neither increased nor decreased since the 1970s . Year-to-year changes in the level of prices are caused by changes in fundamentals, but the range in cotton prices during each season is a reflection of how well the private sector anticipates those fundamentals. If the private sector were doing a better job of anticipating world cotton supply and use, the market-clearing price level would be determined relatively early each season, and prices would tend to that level. However, since the early 1970s, the Cotlook A Index has typically ranged about 15% above and below the average during each season, and there is no empirical evidence that within-season price volatility is decreasing, indicating that forecasts of supply and use by the private sector are not improving systematically. (Forecasts of world cotton supply and use by the private sector may or may not be better than ICAC and USDA forecasts, but price data indicate that private sector forecasts are not improving over time, any more than ICAC and USDA forecasts have.)

The lack of systematic improvement in forecasts of cotton fundamentals is surprising. Thirty years ago, the internet was not available, e-mail did not exist, long distance phone calls were expensive, fax was considered leading edge, and IBM 386’s with 256K of RAM and a port for the “new” 3.5” floppy disks were considered amazing. Weather reporting was slow and dependent on local news services, and weather forecasting was far less sophisticated than today. In the 1980s, people from around the world attended ICAC meetings primarily to listen to country reports from each other because information exchange was still an important ICAC function. In the 1980s, USDA was still conducting what were called Outlook Conferences, the forerunners to the present Outlook Forum. The Outlook conferences had been designed from the 1920’s to bring extension agents from around the country to Washington to review the fundamental outlook for supply and use in each commodity so that extension agents could better inform farmers in their states of the outlook for prices. It was hoped that farmers would then make “rational” production decisions. This format of outlook conferences was necessary because basic information was still a scarce commodity.

As statistician for ICAC in 1987 I sent six surveys to member countries each year via regular mail, including surveys of planting intentions, yield expectations, consumption expectations, estimates of ending stocks, final production, imports and exports, and final estimates of production, consumption, trade and stocks. Once mailed from the ICAC office, typical response times varied from a few weeks at best to three months to never. One of the more glorious achievements of the statistician in 1988 was to send surveys of planting intentions and yield expectations to Switzerland and Hong Kong. Government officials in both countries dutifully returned the survey questionnaires with all blanks filled in with zeros. The survey response from Switzerland included a letter to the new statistician explaining that Switzerland did not produce cotton, and it would not be necessary to send production surveys in the future, as production of cotton was not expected to increase.

Today, with the Internet and all the associated tools of computers, phones, skype, e-mail and web sites, information is available far more quickly on a wider range of data than in the 1980s. Today, we know of weather as it is happening, we know about weekly releases from the China National Reserve, we know of changes in Indian cotton tax policy over night, we know of a strike in the port of Santos or a drought in Australia or a flood in Thailand that has disrupted mill use. We have far more data on monthly trade figures, we know of arrivals in Pakistan and India, we know when the harvest ended in Uzbekistan, we know, we know, we know, many different things.

And, in addition to knowing things, we have far more robust tools to analyze and catalogue, to arrange and compare, to store and to sort, to add and to extrapolate. Thirty years ago, it was still expensive to run a regression, although you no longer had to invert a matrix by hand, but you did need access to a mainframe. Today, with econometric packages built into Excel and robust software packages available, people run regressions without even knowing what a degree of freedom is, much less how regressions are calculated.

However, we don’t know any better today than thirty years ago what it all means in terms of production, consumption, trade and stocks, and ultimately the outlook for prices. Why not?

Economics is a social science

First, it bears emphasizing that economics is a social science. Estimates of world cotton fundamentals represent expectations about the behavior of hundreds of millions of individuals in response to multiple influences, and human behavior is inherently unpredictable. Forecasts of production include imbedded assumptions about farmers’ reactions to changes in output prices, input prices, weather forecasts, labor availability, input availability, storage capacity, marketing opportunities, food security, changes in technology, government policies and other innumerable factors. Forecasts of consumption are really forecasts of textile mill managers’ choices in response to the welter of price information, resource constraints and government policies they face, which are themselves influenced by the lagged impacts of consumer behavior as much as two years earlier. As economists in a highly atomized world, we are depending on offsetting errors in the estimates of behavior of individuals so that forecasts of world totals can be comprehended. No matter how much we try to simplify forecasts into a handful of numbers about world supply and use, we are still trying to predict human behavior, and all the improvements in information technology and analytical techniques in the world will not help us much in doing that.

Shocks Still Happen

Another reason that forecasts have not improved despite enormous improvements in communications and information technology is because we are still shocked by events outside our ability to understand and predict. The near-economic collapse in 2008 and the Great Recession, an overnight imposition of an export ban by India in 2010, and the announcement in March 2011 of an intention to support prices in China through unlimited purchases for the State Reserve, are examples of recent events outside the scope of agricultural analysis used in forecasting supply and demand fundamentals. At both ICAC and USDA, macroeconomic forecasts and the status quo for government policies are taken as a priori assumptions in the development of supply and use forecasts. [Perhaps forecasting agencies should be making greater use of Monte Carlo Simulations of responses to potential economic and policy shocks when discussing the outlook for world cotton supply and use so as to give market participants a better comprehension of the range of possibilities and reduce the focus each month on point estimates.] The ICAC publishes its track records twice a year in hopes that readers will understand that all forecasts include a range of error around each point estimate, and USDA maintains similar data.

Statistical Inputs Are Weak

Nevertheless, even though human behavior is highly variable and unanticipated policy shocks are common, you would still think that after decades of practice and great advances in technology, there might be some improvement in forecasts, but there has not been. Perhaps this is because the information we are getting faster is actually degrading in quality. In other words, the statistics on which forecasts are based are becoming less accurate, thus undermining the value of getting those statistics more easily.

The United States, Europe and the Soviet Union accounted for 38% of world cotton production in 1982/83 but the same countries accounted for just 22% in 2012/13. While there are uncertainties in the estimation of production in all countries, the U.S. has the best agricultural information system in the world, and Europe and the Soviet Union had better information about prospective plantings and potential yields than many countries in Asia and Latin America. [For whatever other ills it may have been responsible for, the Soviet Union and its associated economic grouping, COMECON, was at least good for relatively stable production and consumption and comprehensive statistics if you waited long enough.] Not to disparage information gathering efforts in any countries, but clearly information from India, Pakistan, and China, who now account for a greater share of the world total, is less robust than the agricultural information from the U.S., Europe and the Soviet Union that was available in the 1980s.

As for consumption, the U.S., Europe and the Soviet Union accounted for 34% of world mill use in 1982/83 but just 8% in 2012/13. As with information about production, consumption data from the U.S., Europe and the USSR tended to be of higher reliability than data about mill use from countries like Vietnam, Bangladesh, India, Pakistan and especially China. Consequently, as mill use has shifted from developed countries and the former USSR over the last thirty years, the quality of economic data and agricultural information has degraded.

Developing countries face great difficulties in acquiring data because of highly atomized industries, weak data gathering infrastructure, and fears of taxation and regulation that make survey respondents in many countries wary of participation. There are an estimated 30 million cotton producing households in China, 9 million in India and Pakistan and 3.5 million across Africa . The Cotton Corporation of India publishes guidelines for ginners in 93 languages, and the Cotton Association of Zambia provides extension information in 78 languages. In lieu of income taxes, farmers in many countries are taxed or assessed rural levies based on crop production, so they don’t want to report their production. Many countries levy value added taxes, which naturally discourage accurate reporting of value addition.

Another factor undermining accurate collection of basic data on production and consumption, that might lead to improved understanding and thus improved forecasting, is that government officials in many countries harbor bias against primary industries, especially farming, and they see textiles as a “sunset industry” not to be supported with investments. In many developing countries, farmers are viewed as unproductive, poorly educated, the least ambitious in society, and governments view agriculture as an economic backwater to be diversified out of. Consequently, many governments don’t invest in agricultural information systems because they are just not interested and would rather focus resources elsewhere.

And, finally, there is the problem of just plain old error borne of sloppiness, ignorance or difficulty. As told by Sir Josiah Stamp, Collector of Inland Revenue for the Government of the United Kingdom in the 1920’s, who was recounting a story told by another government official,

“The government [is] extremely fond of amassing great quantities of statistics. These are raised to the nth degree, the cube roots are extracted, and the results are arranged into elaborate and impressive displays. What must be kept ever in mind, however, is that in every case, the figures are first put down by a village watchman, and he puts down anything he damn well pleases.”

For whatever reasons, whether indifference, difficulty, hostility or just incompetence, a greater proportion of cotton is produced and spun in countries with weak systems of data collection today than was the case 30 years ago. Therefore, while our access to data is greater, the data we are gathering is weaker than in years past, and the two factors have approximately offset each other over the last thirty years.

A Call for Transparency

The recent Bali WTO Ministerial declaration with an emphasis on trade facilitation and improved transparency in agricultural industries is a heartening development. Since the 1920’s, governments have tried farmer education, buffer stocks, export and import quotas, trade protectionism, supply management schemes with allotments and quotas, demand enhancement programs, income support and price stabilization programs and input subsidies in efforts to balance supply with demand, to support farm prices and farmer’s incomes and to guarantee food security. But, “transparency” is now being raised at the multilateral level as a formal objective of government agricultural responsibilities. The Bali declaration builds on a G-20 initiative begun in 2011 to explicitly target “transparency” as a tool of improved policy formation and private sector decision making in order to enhance the functioning of the world economy.

In addition, the food supply shocks of 2008 dramatically changed the perception of commodity industries among world leaders. For the first time since the 1930s, commodity industries are seen as positive aspects of the world economy and important avenues for the achievement of Millennium Development Goals, rather than as sectors to be diversified out of. Hopefully, the confluence of increased stature for agriculture within the world economy, combined with a realization that transparency in the form of accurate estimation of production, consumption, trade and stocks in commodity industries is important, will lead to greater government efforts in the next decade at improvement in data collection.

Therefore, I will end on a positive note. First, even though the absolute value of percentage errors in year-ahead forecasts has not fallen, the difficulty in making forecasts has increased with the shift in production and especially consumption. Therefore, the information and technology revolutions have improved forecasting by offsetting the degradation in basic information quality.

Second, as governments concentrate more on the importance of data collection in policy formulation and private sector decision making, and as cotton producing and consuming countries themselves become more developed and have more resources with which to collect data, the basic information that forecasters work with will improve. And, as statistics improve, forecasts will gradually become more accurate.