Transcript State Level Tests of Okun`s Coefficient -

State Level Tests of Okun's
Coefficient -- Implications for
the current U.S. Recession
Okun’s Coefficient
• Key (along with Philip’s curve) macro variable
• Embedded into all sorts of practical models of
the economy
• No theoretical basis – indeed anti theoretical
• It violates the principal of declining marginal
utility
•
A 3% decline in output should result in a 6%
decline in employment, not the opposite
U.S. estimates
• Originally, estimated at 3:1
• Current estimates put it closer to 2:1
• Debate as to whether this reflects a
change in the economy, or is just a better
measurement
International Estimates
• Most European countries have a lower
estimate
• Generally said to be due to labor rigidity
and unionization
• Japan (had) a much higher estimate
• No good estimates for Thailand, due to a
lack of reliable unemployment data
(Bhanupong Nidhiprabha)
Theoretical Issues
• Asymmetries – is Okun’s coefficient
different during upturns and downturns?
• Does Okun’s coefficient change over
time?
• Can it be related to the Phillip’s curve
• What supply and demand (sectoral and
labor) factors influence it?
Measuring Okun’s coefficient
• Yearly vs. Quarterly (with lags) data
• How to detrend
• How is the data gathered, comparisons
across time/polities
• Co-integration, omitted variables, linearity
Differenced equation
• ∆yt = ß0 - ß1 ∆ut + εt
• Where ∆yt was the change in output, ß0 is
the intercept, ß1 ∆ut estimates the change
in unemployment, and εt is an error term.
Gap equations
• yt - yt* = ß0 - ß1(ut - ut*) + εt
• where the star denotes the long run
equilibrium value of the variable.
Expanded out estimates
(Prachowney’s formulation)
• yt - yt* = ά(c –c*) + βγ(l – l*) – βγ(u – u*) +
βδ(h – h*)
• In the above, c is the utilization rate of
capital, l is total employment, u is the rate
of unemployment, and h is hours worked;
in all cases a * indicates the trend variable
Measurement difficulties
• Gap equation estimates rely very much upon the
construction of long run trend variables
• Data needs to be de-trended, both for
seasonality, and for the long and short cycles
• Most papers now use a variety of de-trending
methods
• HP, BK, Arima, BN, other types of Bandpass
filters
Main problem with Okun’s
coefficient papers today
• Okun’s coefficient has become the
plaything of econometricians….
State Level Tests of Okun’s
coefficient
• Will Okun’s coefficient vary between
polities that share a common monetary
policy?
• What factors within the states will cause
the coefficient to vary?
• Can new insights be gained with a new,
large and robust dataset?
Data
• Unemployment was U3 data from the BLE,
1950s for all states, monthly/quarterly/yearly
• Output data was much more difficult to find
• BEA maintains two data sets, the xxxx set, from
1977 (1970 for 26 states) to 1998
•
Approximates GNP, but in many ways is
closer to an income measure
• The xxxx set, from 1998 to 2007 (updating)
which is comparable to GDP measures
Data problems
Unemployment data had no problems
• The output data from 1970 to 1998 had
two major revisions in the method of data
gathering
•
(aside -- how does BEA gather data?)
• Data itself gave some strange results – it
vastly overstated measured/taxable
income
Results (I) 1977-1998
• Differencing gave poor results, unless one
added a dummy variable for 1987
• Then good results, 31 states gave significant
results, somewhat lower then national estimates
• This contradicted Blackley (1990), who got
higher results
• Smaller states gave less significant results, with
much more variance.
• 24 of 25 largest states had significant results,
between .9 and 2.4
Results (II) 1977-1998 (BK
method)
• Gap estimation gave betters results, (42
states), somewhat lower estimates
• Robust to the estimation method used.
• Estimates (generally) ranged between 1.4
and 2, again lower than national estimates
Results (III) 1998 – 2007
• Differencing gave O.K. results (17 of 25
largest states)
• Gap estimates gave poor results (12 of 25
largest states)
• Primarily due to the short data-set, 2 more
years of data should fix this
Implications
• State governments have less ability to use
Okun’s coefficient to reduce
unemployment
• This is especially true for small states
• The smaller the state, the greater the
impact of the national business cycle
• There are still regional differences
Extensions – testing for
asymmetries (1)
• Testing for asymmetries and lags (1977—1997)
• All tests for lags came out negative
• With 50 states, it was possible to test by year
•
Okun’s coefficient was almost always
significant during downturns
•
Much less important during upturns
•
Significant evidence that the coefficient is
asymmetric
•
Aside – risk aversion, threshold effects,
or just clearer data
Extensions – testing for
asymmetries (2)
When the data was split into upturns and
downturns…..
• Okun’s coefficient was consistently larger
in upturns, and smaller in downturns
• Okun’s coefficient was always significant
in downturns, not so in upturns
• Downturns did show lagged effects for one
year
Extension – tests of labor
mobility
• Moran I test – test of long range spatial
relationships
• Ran for 8 regions, and for 48 continental
states
• Regions showed some effect, state level
tests did not
• Similar to results for Spain and Greece
Other Variables
• Used a host of demand and supply
variables
• Taxes, female participation in the labor
force, Age structure, manufacturing base,
etc.
• Many things significant, but few important
• Noteworthy, unemployment insurance was
not important
Size of the state was the most
important variable
• Small states rarely had good results, large
states usually did
Put another way
• California does not care about what
Nevada does, but Nevada cares very
much about what California does.
Extended form of Okun’s
coefficient
• Used the Prachowney method,
theoretically more rigorous
•
But much harder to measure
• As a practical matter, used a reduced form
of it.
• Did not get very good results
Implications for today I
• Okun’s coefficient has been decreasing
coming out of recessions
• Labor markets are more sensitive to
downturns then upturns
• Individual state economies do matter –
some states much harder hit then others
• The ability of an individual state to “grow
out” of a recession is limited
•
Micro policies seem more effective
Implications for today II
• The crash in the housing market could be
impacting labor mobility in a significant way
• Greater disparities between states then in past
recessions
• Role of manufacturing and unions has declined,
role of govt and unions has increased
• Can you achieve growth through investments in
the least productive sectors of the economy?