The Fed has said that it will taper its asset purchases next week if it judges that the labour market outlook has improved “substantially”, and will consider interest rate rises when the unemployment rate falls below 6.5 per cent. Jan Hatzius of Goldman Sachs says that the Fed is suffering from “buyer’s remorse” over the 6.5 per cent threshold, and may reduce it next week. But are they right to use the unemployment rate as their key indicator of labour market conditions? The evidence on this is becoming increasingly murky.
The US unemployment rate has fallen from a peak of 10.0 per cent of the labour force in October 2009 to only 7.3 per cent today, an impressive rate of decline. Yet the employment/population ratio has hardly changed at all over the same period, implying that the whole of the decline in unemployment has been due to a decline in the participation ratio, as disillusioned job seekers have quit the labour force.
On this basis, many Keynesians have argued that the labour market has not improved at all, let alone substantially. The corollary is that the unemployment statistics are giving a seriously misleading indication of the scope for further action to stimulate demand.
This issue has been rumbling along for at least a couple of years, but it came to the fore last week with an important speech by John Williams, President of the San Francisco Fed. He said explicitly that “the preponderance of evidence indicates that the unemployment rate remains the best overall summary statistic”, while “the employment-to-population ratio blurs structural and cyclical influences”.
Mr Williams pointed out that the unemployment rate has consistently been the best single measure of slack in the labour market for many decades, and that it remains very closely correlated with alternative measures of slack derived from other sources. This is supported by the data. The unemployment rate is now about 0.7 standard deviations above its long term mean. The first graph shows three other measures of labour market slack derived from entirely different sources, and all of them are in the region of 0.0-1.0 standard deviations from normal, which substantiates the broad message from the unemployment data.