By

FAO Paper: Premise’s data in Brazil predicts food trends 25 days in advance of official monthly releases

We’re thrilled to share that the United Nations Food and Agriculture Organization (FAO) published a paper anchored by data we captured in Brazil earlier this year. Written by the FAO’s Chief Statistician Pietro Gennari and Senior Statistician Sangita Dubey, and presented at this year’s annual event for stats experts at the International Association for Official Statistics, the 32-page paper compares Premise’s food data in Brazil with the data from Instituto Brasileiro de Geografia e Estatistica (IBGE), Brazil’s official stats-gathering agency.

Among other key takeaways, the FAO determined that Premise’s Brazil Food Staples Index (FSI) can be used to predict food trends 25 days in advance of Brazil’s official monthly releases. During June and July, Premise’s data significantly diverged from the official data – and those price distortions were likely due to the World Cup. Premise’s data revealed an increase in inflation while the government’s data didn’t reflect that until September and October (more on all of this a bit later).

A term called ‘nowcasting’ has been gaining traction in government and economic circles. Coined by Google’s Chief Economist Hal Varian, who also serves as an advisor to Premise, the word essentially means ‘predicting the present.’

For Google, nowcasting has meant shedding light on a wide range of topics, from flu trends to inflation and consumer sentiment. For Premise, it has meant foreseeing food pricing trends 25 days in advance. When it comes to the FAO, whose mandate is the eradication of hunger worldwide, nowcasting can have profound impact on the Organization’s ability to abate food security and food shortage problems in enough time to actually help people, and maybe even save them.

There’s finally a growing consensus among economists and policy-makers that more granular and timely economic statistics are needed as the world economy becomes more complex, interconnected and volatile (below, see Bartelsmans and Doms (2000), Stock and Watson (2002), and Tybout (2000) for examples). This is particularly the case in developing economies, where supply-chain and infrastructure fragility leads to massive price volatility in food and other critical staples.

Better, faster, more reliable data means greater risk management. A 2014 World Bank report mentioned in the paper attributes food riots to sharp food price increases. In recent months, the Ebola outbreak in West Africa has led to massive price spikes on food staples and a highly complex economic effect on the markets, something we’ve found to be especially true in Liberia.

Let’s dig into the data mentioned in this paper and how we capture it.

Equipped with Android phones, contributors to the Premise network take photographs of food items based on a sampling frame designed to cover a broad range of neighborhoods and store types, as well as the consumption figures quoted by official sources.

The people in our network are at the heart of what we do – and it’s because of them that we can get a real-time read on food staples inflation and other key indicators across 23 key markets in Brazil. In other words, human intelligence is part and parcel to gathering such granular data. Their data is then uploaded to the Premise platform, classified and analyzed, and distilled into actionable data and insights.

Getting into some of the nitty gritty numbers: for the time period May 2013 to August 2014, the Premise daily FSI conveys substantially similar information as compared to standard governmental releases:

“…10 days before month-end and up to 25 days before the official IBGE release. Indeed, the later analysis will show that the first 7-day average of Premise’s daily FSI, available before mid-month, does a reasonably good job of predicting or now-casting the monthly food-at-home CPI.”

In terms of Premise’s data stacking up against the official inflation data, the graph below shows a comparison of month-over-month inflation from June 2013 to Aug 2014: Premise daily FSI converted to monthly average values at four different smoothing intervals; Instituto Brasileiro de Geografia e Estatistica (IBGE) official food CPI data.

Figure 1: Premise daily FSI vs. IBGE, June 2013-August 2014 Premise daily FSI vs. IBGE

With exception of July and August 2014, official data and the Premise series all have similar movements. Now, look more closely at July and August and things get more interesting:

“Between September 2013 and January 2014, it almost appears that official data lag Premise series. This interpretation would not be valid, however, given that both data sets set out to measure the same phenomenon. Most problematic, however, is the July and August 2014 data, in which official statistics show a fall in food prices, while Premise data shows an increase.”

In terms of mean absolute percentage error (MAPE), the FAO researchers found that Premise could be used to nowcast the official IBGE data accurately up to 25 days before the official monthly releases during the period from April to June.

As you’ll see below, there were significant divergences in July and August. In June and July Premise began tracking significant price distortions likely due to the World Cup. These effects lingered well into August due to price-stickiness.

Figure 2: Mean absolute percentage error, Premise vs. IBGE Mean absolute percentage error, Premise vs. IBGE

All of this gets to the heart of why we started this company in the first place. We knew there had to be a better, faster and more reliable way to capture important economic data. This is validation of the progress we’ve made – and a harbinger of what’s to come.

Read the full paper at the 2014 IAOS conference Now-casting Food Consumer Price Indexes with Big Data: Public-Private Complementarities (slides).

Explore our data: data.premise.com

References

Bartelsmans, E. J and Doms, M. 2000. “Understanding Productivity: Lessons from Longitudinal Microdata.” Journal of Economic Literature, 38, 3.

Stock, J. H and Watson, M. W. 2002. “Macroeconomic Forecasting Using Diffusion Indexes.” Journal of Business and Economics Statistics, 20, 2.

Tybout, J. R. 2000. “Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?.” Journal of Economic Literature, 38, 1.