Friday, June 09, 2017

Industrial production fell by 0.8% over the year to April, with particularly sharp falls in the energy and consumer non-durables sectors. This follows forecasts of such a fall on this site in recent months. While the political landscape is now particularly uncertain, forecasting is more than usually hazardous, but the predictions of my neural network model for this series over the next 24 months are reported below.

Thursday, May 11, 2017

Data on industrial production indicate that output growth in this sector slowed in March to 1.4% year on year. This follows sharp declines in each of the previous two months, and on a monthly basis industrial output is now some 1.9% lower than its peak in December of last year. Since there was also a mini-peak in the series in April 2016, it would not be surprising if the year-on-year series were to turn negative next month.

The series, along with predictions for the coming two years from my neural network forecaster, appears in the graph below. As ever, forecasts should be treated with caution, not least given the present political uncertainties.

Friday, March 10, 2017

Following the substantial uptick in manufacturing output fuelled by sterling's depreciation at the end of last year, the January data indicate a month-on-month fall of some 0.9%. This has contributed to a month-on-month fall of 0.4% in total industrial output. Year-on-year, industrial output still shows a large rises, of some 3.2%.

The slowing of output in January leads to another major revision in my neural network forecast for this variable over the coming 24 months - illustrating again that forecasting in such volatile times is a hazardous activity. The latest forecast is shown below - and is clearly more consistent with forecasts produced over the course of most of last year than with the one produced last month.

Friday, February 10, 2017

The latest statistics on industrial production indicate that, compared with a year earlier, output in the production sector in December 2016 had grown by some 1.2%. This spurt of growth is new. Indeed, industrial output grew by over 3.1% over the last 2 months of 2016, following some earlier reverses. The main driver of this growth is in the manufacturing sector, which, over the course of the year, increased output by some 4 per cent. Growth in manufacturing since October has been particularly strong - at 3.5 per cent over the two months alone.

Using these data to update my neural network forecaster for industrial output means that - with the positive annual growth rates recorded in each of November and December of last year - the forecast is now for continued growth over the period to the end of 2018. The depreciation of sterling has clearly given manufacturing exports a boost, and while the series dipped in October of last year this dip has proved to be much milder and shorter-lived than anticipated. The uncertainties brought on by Brexit have clearly made forecasting an even more hazardous activity than usual!

Friday, January 06, 2017

Comments by Andy Haldane, chief economist at the Bank of England, comparing economic forecasts to the famous failure of Michael Fish to predict the October 1987 hurricane have been seized upon by the media. The relevant part of Haldane's commentary comes in the 5 minutes from 15m30s in this video.

A number of points are worth making about this. First, the specific forecasting failure that Haldane compares to Fish is that of the financial crash leading to the Great Recession. Some media outlets have suggested otherwise. Haldane does comment on the Bank's forecasts for the post-referendum period and notes that the economy has been more resilient so far than had been expected, but he continues to expect a relatively tough year in 2017.

Focusing then on the major forecasting failure in 2008, he identifies two contributory factors. The first (extending the analogy with meteorology) is a lack of data. With better data, better forecasts can be produced. The second is arguably more fundamental. As Haldane notes, the forecasting models tend to work well when the economy is close to equilibrium, but perform badly during the (more interesting) periods following a shock. They clearly need to be redesigned, and indeed are being redesigned, better to accommodate such extreme events. Much effort since the crisis has gone into developing macroeconomic models to include imperfectly operating housing markets, and it is likely that this effort will contribute to more successful forecasting in future.

That said, economies are made up of people with free wills, and forecasting in this context can never become an exact science. The forecaster's tools - be they VAR models, neural networks, DSGE models or whatever - allow the evidence to be marshalled systematically in order to produce informed estimates of the likely time paths of key economic variables. But they are informed only by what is known at the time of the forecast, not magically informed by data that are unavailable. That said, data on the vulnerability of the sub-prime sector were available in 2007, and it is certainly fair to say that these should have been given greater heed in forecasts.

However, while many laypeople consider forecasting to be a major part of what economics is all about, that perception is misleading. Most economics is based on generating hypotheses that are then tested on historical data. This allows some stylised facts to be determined, and helps us understand a complex world - for instance: production quotas raise the price of oil; or restricting trade is harmful to growth. The body of economic understanding that has developed in this way over many years is in no way challenged by the fact that (in common with everybody else) economists lack perfect crystal balls.