Reading a food price forecast: what the numbers can and cannot tell you

Every December, a forecast of next year's food prices makes the evening news. It usually arrives as two numbers, a percentage range and a dollar figure for a family of four, and by the next morning both have been rounded into a headline. A few months later, Statistics Canada's monthly inflation figures start arriving, and people check one against the other.

That comparison is reasonable, but it only works if you know what each number measures. A forecast and the official index are built differently, cover slightly different things and answer different questions. This piece takes a forecast apart in the order a careful reader would: what is being predicted, how the prediction is made, how it went in the last two years, and how to apply it to your own grocery bill. It offers no forecast of its own.

Step 1: find out what is being predicted

Statistics Canada measures consumer prices with the Consumer Price Index, a fixed basket of goods and services priced every month. Food is one of the major components, and it splits into two parts that behave quite differently:

• Food purchased from stores: groceries, from meat and dairy to bakery products and fresh vegetables.
• Food purchased from restaurants: meals and snacks bought from table service, fast food and cafeterias.

Each part has a weight based on how much households spend on it. In the basket introduced in June 2023, which uses 2022 spending patterns, food as a whole made up roughly a sixth of the index. The mechanics of the basket and its weights are explained in how food inflation is measured and forecast in Canada.

The first question to ask of any forecast, then, is which of these it predicts. Is it store food only, or store and restaurant food together? Is it the average change over the whole year, or the change from one December to the next? Those choices can move the figure by a couple of percentage points with no change in the underlying prices.

Annual average versus year over year

Statistics Canada reports both. The year-over-year figure compares one month with the same month a year earlier. The annual average compares the average level of prices across one calendar year with the average across the year before. When prices rise quickly early in a year and then flatten, the two can diverge widely. Annual forecasts usually refer to the annual average, while the monthly news usually reports the year-over-year figure for the latest month. Comparing a forecast with a single month's headline is comparing two different things.

Step 2: know who makes the forecast and how

The best known Canadian forecast is Canada's Food Price Report, an annual report produced by researchers at Dalhousie University, which leads the project, the University of Guelph, the University of Saskatchewan and the University of British Columbia. It is published in early December and gives a range for overall food prices, ranges for categories such as meat, dairy, bakery and vegetables, and an estimate of annual food spending for a reference family of four.

According to the universities' own descriptions, the forecast combines historical price data with machine-learning models and other predictive tools developed over several years, along with added variables such as expected general inflation and the exchange rate. That is typical of the field. Banks, consultancies and government economists build similar models, and most share three ingredients:

1. Momentum. Food prices have long memories. Contracts between suppliers and retailers are often renegotiated once or twice a year, so increases already agreed will show up on shelves for months.
2. Inputs. Grain, oilseed and livestock prices, fertilizer and fuel, packaging, wages, and transport all feed into the cost of food. Some of these are traded on futures markets, which give a rough guide to expectations.
3. The dollar. Canada imports a large share of its fresh fruit and vegetables, many processed foods and some meat, mostly priced in US dollars. A weaker loonie makes imports dearer.

Then come the judgment calls: whether a drought will persist, whether a disease outbreak in livestock will spread, whether retail competition will tighten. These are the parts no model handles well.

Step 3: check the record, without hindsight bias

The last two years show both what forecasts can and cannot do. The figures below come from Statistics Canada's annual Consumer Price Index reviews and the universities' published reports.

In 2022 the forecast was the highest the report had ever predicted, and it still fell well short of what happened on grocery shelves. The forecast was published in December 2021, before Russia's invasion of Ukraine in February 2022 pushed up the prices of grain, vegetable oil, fertilizer and energy. Store food prices rose 11.4 per cent year over year in September 2022, the fastest pace since August 1981.

In 2023 the forecast range was closer, but store prices still rose more than its upper end, led by edible fats and oils (+16.9 per cent), preserved fruit (+11.5 per cent), bakery products (+10.7 per cent) and cereal products (+10.0 per cent). Restaurant prices rose a little more than that edition's separate restaurant range of 4 to 6 per cent.

By early 2024 the picture had changed again. Store food prices in March 2024 were 1.9 per cent higher than a year earlier, the smallest year-over-year increase since July 2021.

None of this proves forecasting is pointless. It shows that forecasts are good at carrying momentum forward and poor at anticipating shocks, which is what you would expect. The December 2023 edition, published for 2024, projected overall food prices rising 2.5 to 4.5 per cent. Read in that light, it is a statement of where prices are likely to go if no large shock arrives.

Step 4: read the range, not the midpoint

A forecast range is not a promise that the outcome will fall inside it. It is the forecaster's way of saying how uncertain the model is. Most published food price ranges are about two percentage points wide, which sounds precise until you look at the last two years.

A useful habit is to ask three things of any range:

• What would put the outcome at the top? Usually a weaker dollar, a poor harvest somewhere important, or higher energy costs.
• What would put it at the bottom? Usually easing input costs, a stronger dollar, or more aggressive discounting by retailers.
• What would push it outside the range altogether? A war, a pandemic, a major disease outbreak or a trade dispute that closes a market.

If the honest answer to the third question is "quite a few things", the range is a central case, not a boundary.

Step 5: spending is not the same as prices

The family-of-four figure is the part people remember, and it causes the most confusion. The December 2022 edition estimated that a family of four would spend about $16,288 on food in 2023. The December 2023 edition put the 2024 figure at $16,297.20, an increase of up to $701.79, which implies that the family's actual 2023 spending came in near $15,600, several hundred dollars below the earlier projection.

Yet prices in 2023 rose faster than forecast. How can both be true? Because spending reflects behaviour as well as prices. The researchers read the gap as a sign that households bought less food, traded down to cheaper products or switched brands. Price indexes are designed to hold the basket steady. Household spending does not hold anything steady.

A worked example: applying a forecast to one household

The numbers here are illustrative and describe no real household.

Suppose a household spends $1,000 a month at grocery stores and $250 a month at restaurants, a total of $15,000 a year. It wants to know what the 2024 forecast range of 2.5 to 4.5 per cent might mean.

• At 2.5 per cent, the same food would cost about $15,375, or $375 more over the year, roughly $31 a month.
• At 4.5 per cent, it would cost about $15,675, or $675 more, roughly $56 a month.

Now adjust for the household's own basket. Suppose it spends heavily on meat and bakery products, which the report expected to rise faster than average (5 to 7 per cent), and little on dairy and fruit, which were expected to rise slowest (1 to 2 per cent). Its personal increase could sit above the headline range even if the national forecast proves exactly right. A household that eats mostly fruit, dairy and home-cooked staples could sit below it.

Then add behaviour. If the household switches to store brands for a third of its grocery spending and saves, say, 15 per cent on those items, it cuts about $600 a year from its grocery bill. That one change outweighs the entire difference between the top and bottom of the forecast range. This is why the forecast tells a household roughly how hard the wind will blow, not how much it will end up paying.

What the forecast is good for

Used well, a food price forecast is a tool for planning, not for predicting a specific receipt. Grocers and food processors use forecasts to set budgets and negotiate supply contracts. Food banks and municipal planners use them to estimate costs for the coming year. Households can use them to sense direction: whether price increases are likely to speed up or slow down, and which categories are under the most pressure.

What a forecast cannot do is see around corners. The most useful single habit is to note what assumptions the forecaster made about the exchange rate, energy prices and harvests, then watch those three things through the year. When one of them moves sharply, the forecast is out of date, whatever its range said. For the price differences that come from crossing a border rather than waiting a year, see whether food is really dearer in Canada than in the United States, and for more on how supply-managed foods such as milk are priced, see how Canada's dairy quota system works. Further pieces on the checkout bill are on the food prices issue page.