One of the conundrums of interest rate forecasting is that yields are driven by the expectations of central bank policy. Central bank policy is driven by some form of reaction function to inflation and labour market slack. But what happens if your forecast for inflation (or the economy more generally) differs from that of the central bank? Well, when your bond positions are marked to market, you would lose money if your forecast for the tendency in the policy rate is wrong. Therefore, you need to understand what the central bank will do, even if you believe what they are doing constitutes a policy error. Hence the intense focus on dissecting the missives from central bankers, bringing to mind the days of Kremlinology.
Because I do not give investment recommendations to random strangers on the internet for very good legal reasons, I am freed from the necessity to obsess over what the Fed is really thinking. Instead, I can focus on what I think is happening, without worrying whether the Fed is marching in a different direction.
But I would point out that there is limited chance that the Fed would even think about hiking rates before some time in the "middle" of 2015 (I use a pretty wide definition of "middle"). Therefore, the Fed does not have to make any decision of any import for quite some time, and so they are free to hint whatever they want to market participants, who unfortunately have to get into positions months ahead of the turning point. There is a lot of data that will arrive ahead of that date, making the interpretation of the data currently available moot.
What Is The LMCI?
In Assessing the Change in Labor Market Conditions, Hess Chung, Bruce Fallick, Christopher Nekarda, David Ratner (link given above), the LMCI is described as follows:
The U.S. labor market is large and multifaceted. Often-cited indicators, such as the unemployment rate or payroll employment, measure a particular dimension of labor market activity, and it is not uncommon for different indicators to send conflicting signals about labor market conditions. Accordingly, analysts typically look at many indicators when attempting to gauge labor market improvement. However, it is often difficult to know how to weigh signals from various indicators. Statistical models can be useful to such efforts because they provide a way to summarize information from several indicators. This Note describes a dynamic factor model of labor market indicators that we have developed recently, which we call the labor market conditions index (LMCI). Details of the data, model, and estimation will be presented in a forthcoming FEDS working paper.The LMCI is a factor model which summarises the central tendency of a large number of underlying time series. I discussed another such model, the Chicago Fed National Activity Index, in an another article on aggregate indicators. For those of you more familiar with fixed income modelling, it is the first principle component of some related time series in Principle Component Analysis (PCA) of the yield curve.
There are 19 variables used to construct the LMCI, which I list below (sourced from Table 1 in the linked article; see that table for more details):
- Unemployment rate
- Labor* force participation rate
- Part time for economic reasons
- Private payroll employment
- Government payroll employment
- Temporary help employment
- Average weekly hours (production)
- Average weekly hours of persons at work
- Average hourly earnings (production)
- Composite help-wanted index
- Hiring rate
- Transition rate from unemployment to employment
- Insured unemployment rate
- Job losers unemployed less than 5 weeks
- Quit rate
- Job leavers unemployed less than 5 weeks
- Jobs plentiful v. hard to get
- Hiring plans
- Jobs hard to fill
As a result, I have no argument against the basic principle. I would have to replicate the work - which I currently do not have the time to do - to look at the particular decisions made by those authors to verify that the construction was not slanted in one direction or another. My guess is that there would have been considerable pressure from both hawks and doves to keep the indicator from being too biased in any direction.
Wage Stickiness - Another Piece Of The Puzzle
Janet Yellen noted that wage stickiness - the psychological avoidance of wage cuts - can also explain some of the apparent anomalies in labour market data. One worry of the doves is that there is "pent up deflation" in the system - since businesses were unable to cut wages, they will make up for that by not raising them as much as they normally would during an expansion.
I showed how incorporating wage stickiness into a NAIRU-based wage inflation model improved results in an earlier article. I imagine a similar model could use the LMCI to create an improved (wage) inflation model, which would avoid the problems of just using the Unemployment Rate as the measure of labour market slack.
Is There Only One Labour Market?
Aggregating myriad transactions into a composite measure is a necessity to do sensible macro-economics; for example, many Austrian economists reject the possibility of aggregation, and they are left with a theory that says very little (government is bad because it is assumed to be so). It is straightforward to say that if it was possible to split aggregated variables into sub-sectors, an economic model would be better. In particular, it would be great if we could split the labour market into hundreds of markets based on occupation. Unfortunately, we cannot hope to fit such a model to data. (One can use the output of a macro model to make predictions about sub-sectors, which is a different task.)
Therefore, I am cautious about suggesting that aggregated models are inherently bad because they are working with aggregated data. But in this case, there seems to be very good evidence that the labour market in the United States is highly polarised. Splitting the labour market into "high skill" and "low skill" segments would make the model much more realistic. In order to fit the model to data, a somewhat arbitrary division of industrial sectors between the "low skill" and "high skill" segments would be needed.
I doubt that I have the resources to single-handedly develop such a model. But it appears that it would explain much better the facts on the ground. "Low skill" workers have no bargaining power, but nominal rigidities keep wages from falling in this sector. Meanwhile, the "high skill" segment of the labour market is relatively tight, and this keeps the overall inflation rate away from zero.
Going forward, one could imagine scenarios where the predictions of such a model that would diverge from aggregated labour market models (such as those that would use the LMCI):
- If job growth in the "high skilled" segment slowed, it would be possible for there to be strong aggregate job growth without inflationary implications, as the "low skilled" segment catches up.
- If job growth in the "high skilled" segment picks up, it might be possible for inflation to rise further, even if aggregate job growth is tepid.
* My spell-checker hates when I jump between American and English/Canadian spellings. Since these are the proper names of American economic series, I leave the spelling of labour as "labor".
** Apparently I missed having a retrospective of the first year of this website's life.
(c) Brian Romanchuk 2014