The costly consequences of dodgy data

Without a shadow of doubt, the most frustrating aspect of economic forecasting is the fallibility of the raw statistics on which we work. Users of macro-economic forecasts may regard this complaint as an excuse for poor performance, but it is more than that. Before even starting to prepare a forecast, it is essential to come to a view about which aspects of recent economic data should simply be disbelieved. For to take everything at face value can often prove to be a huge mistake.

In the last two meetings of the Treasury's independent forecasting panel, more time has been spent discussing the potential inaccuracies of recent data than arguing about the stance of fiscal or monetary policy. Last July, for example, the panel was puzzled by the strength of manufacturing output, which appeared to be growing at an annual rate of 6-8 per cent at the time.

In the panel's report, several possible explanations for this behaviour were proposed. Some members felt that export volume or stockbuilding might have been growing more strongly than was then shown in the national income figures. Some were inclined to dismiss the figures completely, arguing that they would probably be revised downwards. Others, though, felt that the figures provided encouraging evidence of a supply- led recovery following the sterling devaluation last autumn.

It now appears that the sceptics were right, albeit for a reason that none of them guessed at the time. In last month's figures, the path for manufacturing output was duly revised downwards by almost 2 per cent, thus reducing the annualised growth rate in the first half of this year from 6-8 per cent to a much more normal 3.5 per cent.

And why did this revision take place? Mainly, we are told, because the Central Statistical Office for the first time used an export price series to translate the nominal value of manufactured exports into volume terms. Previously, they had apparently been dividing the value of exports by the selling price of domestic manufactures to calculate the volume series.

With export prices rising much more rapidly than domestic prices this year, this change substantially reduced the estimated level of manufacturing output. There was much mirth in the forecasting panel meeting last week when we realised that none of us had remotely suspected that the CSO might have been using an inappropriate price series in the first place. So much for the supply side miracle]

This is not to say that the government statisticians have a simple task. A modern economy is not an easy phenomenon to capture in figures, especially when the structure of output is changing rapidly, when inflation is varying wildly, and when high rates of value added tax give the private sector a monetary incentive to disguise the true level of its transactions.

Estimates published last month indicate that real gross domestic product in 1987 and 1988 grew by 4.5-5 per cent in both years. At the time, the CSO thought that the growth rate was a much more manageable 3 per cent, which is one reason why Lord Lawson saw no cause to tighten monetary policy until it was far too late.

Similarly, for much of 1992, the CSO denied that any recovery in output was taking place at all. Even as late as the spring of this year, the various 'Golden Guru' contests for 1992 bestowed their prizes on forecasters who had said that the 'green shoots' were illusory. But the latest estimates for non-oil GDP show that the recession bottomed in the first quarter of 1992, which means that the recovery has already lasted as long as the recession did.

Once there have been further upward revisions to the data, a re-run of last year's forecasting awards might look a little different. And more generally there seems little point in trying to compare forecasting track records when the variation between the different forecasters is often smaller than the revisions to the data they are forecasting.

By no means all of this is the CSO's fault, but some of the problems posed by the statisticians in the past have been entirely of their own making. For example, for much of the late 1980s the CSO published three separate estimates of GDP (or four if we count the average of the other three), and these frequently bore not the least resemblance to each other.

Fair enough, there are indeed three different ways of estimating GDP, and these routinely produce different results. But in all other countries, the statistical services have always seen it as their job to produce their best estimate at any given moment of how GDP is performing, even if this means that they have to force the different measures to agree with each other. This is also what the CSO now does, and not before time. But the confusion caused by different estimates in the late 1980s almost certainly contributed to the policy mistakes of the period.

The burning statistical issue of the present moment concerns the trade figures. Since January of this year, the introduction of the single market within the European Community has meant that the CSO can no longer count EC imports and exports as they pass through the ports. So trade with the Community is now measured by sample surveys of large companies.

This is supposed to cover 97 per cent of all trade, so there should be no problem. However, as the graphs show, both import volume and export volume have collapsed in the case of EC trade, while neither has dropped very much in the case of trade with the rest of the world (where the old statistical method is still used).

There seems to be no valid reason to explain the decline in imports from EC countries, so it seems reasonable to conclude that total imports have been substantially understated so far this year. But there is a possible reason for the pattern of exports - European markets have been weaker than world markets, so part of the drop in exports to the EC might well be genuine.

A sensible guess might be that both imports and exports have been under-recorded, but the former by less than the latter. Consequently, the trade figures may recently have been flattering to deceive.

But this is only a guess - which in turn means that forecasts for the trade figures, and the components of GDP, must be based on guesswork as well.

Of course, all of this trouble could have been avoided if the Government had allowed the CSO to run the new statistical methods alongside the old for a trial period last year, thus permitting comparisons to be made.

But this was deemed too much of a 'burden' on British industry. The much greater burden that could later fall on industry if serious policy errors are yet again caused by ropey economic data is, of course, conveniently forgotten.

(Graphs omitted)