How to read economic figures

What figures do statistical and economic agencies use to describe trends, and what are the peculiarities and pitfalls? In this publication, we give some pointers to better interpret economic figures.

This article will talk about:

To identify trends, you look at the development of an indicator between two moments. But exactly which two moments those are can influence your conclusion considerably. To illustrate, take the fictitious unemployment figures in Figure 1. The most recent figures show that in May of year 2, more than 310,000 people in the Netherlands were looking for paid work. In the same month a year earlier, the figure was 270,000. Judging from this comparison, it is tempting to argue that unemployment is on the rise in Year 2: after all, in that month the Netherlands had 15 percent more unemployment than in Year 1.

Yet that is not the correct conclusion. In fact, Figure 2 shows the intermediate monthly figures. These show that there are indeed more Dutch people unemployed in May of Year 2 than in May a year earlier, but that unemployment is no longer increasing at all but has actually been decreasing for four months in a row. In general, the further apart the moments you compare are, the more likely you are to overlook current developments.

So anyone who wants to know as quickly as possible about changes in the economy prefers to compare today with yesterday, or this week with the week before. An acceleration, deceleration or reversal in the trend can then be seen as quickly as possible. However, in practice, most economic figures are published no more often than once a month. Examples are statistics on unemployment, inflation and house prices. Some figures, such as gross domestic product (GDP), are published at most every quarter in the Netherlands and many other European countries. But the principle remains the same: to be as close to current events as possible, it is best to compare the figures of the most recent month with those of the month before (month-on-month, also called: MOM, m-o-m, or m/m). Or those of the last quarter with a quarter earlier (quarter-on-quarter, also called: QOQ, q-o-q, or q/q). Note though that it is premature to speak of a trend based on a single monthly or quarterly figure. This is because those figures might be driven by random and/or one-time events. The shorter the period, the more likely this is the case. Therefore, several consecutive periods pointing in the same direction are needed.

Sometimes there are good reasons choosing to compare with the same period in the previous year. The most important is the presence of seasonal effects (see also 'Beware of holidays and seasonal effects’).

Comparing with a month or quarter earlier is not always the best option. Figure 4, for example, shows the turnover figures of hotels and restaurants in the Netherlands in three consecutive years. It is immediately noticeable that turnover depends strongly on the season: the hospitality industry is, not surprisingly, a lot busier every year in spring and summer than in winter. Therefore, based on a comparison between the second quarter (spring) and the first quarter (winter), it is difficult to determine the underlying trend.

To be able to say something about this, it would in this case be better to compare the same seasons with each other and thus to compare with a year earlier (year-on-year, also called: YOY, y-o-y, or y/y). For example, by comparing the figures of the third quarter of Year 3 with those of the third quarter of Year 2. This shows that the hotel industry had a better summer in Year 3 than in Year 2. Unfortunately this year-on-year comparison brings up the disadvantage described earlier: with a y-o-y comparison you can overlook current trends.

Fortunately, there are methods to account for seasonal patterns, such that it is still possible to do a month-on-month or quarter-on-quarter comparison, despite a winter dip or summer peak. Many statistical agencies publish such seasonally-adjusted figures for certain series, including for unemployment, economic growth and also activity in the Dutch hospitality industry – shown in Figure 4. This shows faster than Figure 3 that the hospitality industry in the Netherlands grew in the three selected years.

In addition to seasonal patterns, some economic series also suffer from working-day, shopping-day, and holiday effects. Not every month has the same number of days, and months with more days provide additional opportunity for households to consume and for businesses to produce. Moreover, some months have more weekend days than others, and a Saturday in a sector such as the hospitality industry is different from a Tuesday. Furthermore, holidays such as Easter fall in March one year and April the next. And hotel stays, for example, are usually more expensive during holidays and vacations.

Many economic series involve seasonal effects, workday effects, holiday effects, and/or shopping-day effects to a greater or lesser extent. A few examples are unemployment, household spending, home sales, consumer prices (inflation) and the size of the economy. Statistics agencies make some of these series available with a seasonal and/or working day and/or buying day adjustment. In some cases, we do this ourselves with the help of statistical methods. However, seasonal corrections are never quite perfect, therefore it remains true that based on just one monthly or quarterly figure it is too early to speak of a trend.

A term occasionally used in our publications is ‘carry-over’. For example, to indicate that a year's economic growth is down compared to the previous year, but that the annual figure is still high due to carry-over from last year. It’s a quirk that follows from the standard way of reporting annual figures, and that is to compare the average (or total) of one year with the average (or total) of the year before. As a result, a trend in one year often affects the following year's annual figure as well – even if that trend has since died out (or even reversed).

Take the fictitious unemployment figures in Figure 5 as an example. More people were unemployed in Year 2 than the year before, but that does not automatically mean that unemployment also rose in Year 2. In fact, the underlying quarterly figures in Figure 6 show that unemployment rose only in Year 1, from 240,000 people in the first quarter to 330,000 in the fourth quarter. But because the annual figure is the average of the four quarters in a year, that figure is still a lot lower for Year 1 than the figure for the last quarter of Year 1, at 285,000. The Year 2 average is 330,000 because the number of unemployed people remained at 330,000 each quarter. So the 45,000 people gap between the two annual figures is not caused by growth in Year 2 but by a positive carry-over. Where the carry-over effect is the difference between the last value in Year 1 (330,000) and the average of Year 1 (285,000).

So annual figures only show that unemployment was higher on average in Year 2. Without quarterly, monthly or weekly figures, little can be said about underlying trends. Moreover, due to the carry-over effect, the unemployment rate in Year 2 in this example would have to fall very sharply to arrive at an annual figure lower than in Year 1. For this reason, our reports sometimes show that an annual figure is still higher than that of the previous year despite a downward trend in the year itself.

Carry-over sometimes makes it virtually impossible to say anything about any underlying trends based on a comparison between annual figures. For example, Figure 7 shows that from Year 1, there is a negative carry-over of 275,000 (the latest value) minus 312,500 (the average value) = -37,500. This makes the average for Year 2 lower than in Year 1, even though unemployment rises sharply again in Year 2.

Many economic figures are colored by price changes. For example, sales in the hospitality industry may increase because more drinks are sold, or because the price of a glass of beer has gone up. Only the former says something about how much is being made and sold. And that is precisely what economists are often interested in, because production and sales say something about employment and productivity, and all sorts of related things.

Therefore, when looking at economic figures, it is common to look at so-called volumes. Volumes give an indication of the number of products and services made or purchased and/or their quality. Volume figures are adjusted for price increases or decreases in a product or service of the same quality. Higher-quality products are often more expensive, but there is no need to adjust for that because higher quality is counted as more volume as it is assumed that higher quality products and services entail more production (see also "Comparing apples to oranges through composition effects"). Think of a compact car versus a more expensive SUV, a brandless handbag versus a designer handbag, or a beer versus a pricier cocktail.

Figure 8 shows activity in the hospitality industry compared to a year earlier, with and without the influence of prices. In values, not adjusted for price changes, hotel activity is more than 2 percent higher than the year before. But in volumes, i.e., adjusted for price changes, activity is almost 2 percent lower. This may indicate that hotels had fewer lodgers (or that guests booked fewer upscale hotels) but that room rates increased.

For restaurants, it can be seen that activity is considerably lower in both value and volume terms than a year earlier. The difference between the two is an indication that prices were higher, but that there were fewer customers, they bought less and/or chose cheaper snacks and drinks. For example, because they preferred simple beers or soft drinks to fancy cocktails that year.

In our growth forecasts for the Dutch economy, we are talking (unless otherwise stated) about the development of the size of the economy in volumes, i.e., net of price changes. This is also sometimes called the development 'in constant prices' or 'in real terms'. When no price changes are taken into account, we refer to the development 'in real prices' or 'in nominal terms'.

In the beginning of this publication, we wrote that to chart trends, you look at the development of an indicator between two moments. But sometimes you unintentionally compare apples with oranges. This is because what a time series measures is not always exactly the same every moment, which can lead to so-called composition effects. The average selling price of houses is a good example of this. This statistic shows how much was paid on average for houses that changed owners in a given period. That sounds straightforward. But Figure 9 shows that the mix of houses sold sometimes has a relatively large number of apartments, while at other times there are many detached houses. So a difference in the average sales price between the two times says more about the composition of houses sold (smaller, mostly cheaper apartments versus larger, often more expensive detached houses) than the underlying trend in owner-occupied house prices.

Moreover, in the case of houses, many more such composition effects can be imagined: expensive city versus cheaper city, desirable neighborhood versus less popular neighborhood. Large corner house versus small corner house. Detached house with backyard facing northeast versus detached house with a garden facing southwest. Pre-war townhouse or 1990s terrace. All of these factors play a role in the price paid for a house, besides the things that affect the overall level of house prices, such as rising or falling mortgage rates.

The sales price of houses is not the only time series that suffers from composition effects. Do all Dutch people work fewer hours, or do Dutch people work fewer hours on average because of the rising share of people over the age of 55 in the labor market – with people over 55 often working part-time? Are Dutch people spending more on cars because manufacturers have raised their prices, or because they are buying more expensive, electric versions? Compositional effects are not always a problem: sometimes the change in composition itself is of interest. For example, the size of the Dutch economy has grown strongly over time precisely because its composition has changed.

But anyone looking for trends would do well to look critically at exactly what an indicator measures, and therefore what an increase or decrease does and does not say. Fortunately, statistical agencies help with this: for example, in addition to the average sales price, Statistics Netherlands and the Dutch Land Registry publish the price index of existing owner-occupied homes, a figure which is ‘cleaned’ to reduce common composition effects. For other time series, sub-indicators are often available that allow you to see for yourself whether composition effects may play a role, such as labor force participation and the share of full- and part-time workers by age group. Correcting for all potential composition effects is, unfortunately, virtually impossible, so caution is still required when drawing conclusions.