Transcript Slide 1

Extracting the Cyclical
Component from Australian
Multi-Factor Productivity
Mark Zhang
Lewis Conn
1
Background
The Australian Bureau of Statistics (ABS) produces two
measures of Multi-Factor Productivity (MFP) growth
 Growth between adjacent years
 Average annual growth between productivity peaks
The later is a more consistent measure of productivity growth
Reserve Bank Statement on Monetary Policy (9th Nov 2006)
"... Australia’s economic expansion has now reached a mature stage in
which previously unused productive resources have been substantially reemployed ... this combination suggests that there may have been some
underlying slowdown in productivity, either of a cyclical or structural nature,
though its extent is difficult to explain. "
2
Deriving Average Annual MFP Growth
In theory, at MFP peaks there is full capacity utilisation
By measuring MFP growth from peak to peak we assume
we have consistent capacity utilisation
MFP Peaks are currently derived using an 11 term
Henderson time series filter (Aspen, 1989)
Separates the business cycle from long term trend
Other economic series (Labour market, business
expectations) are also considered when declaring peaks
3
Objectives of This Study
 Improve analysis and understanding of trends in
multi-factor productivity
 Review, update and explain the choice of method for
estimating peaks in the productivity series.
 Analysis of impact of methods on the productivity series
 Analyse how industries' contribute to the aggregate
productivity cycle (Phase 2)
 Investigate how capacity utilisation may be taken into
account when comparing productivity peaks (Phase3)
4
Filters Considered
Current ABS Method
11-Term Henderson (1916) - Linear low pass filter
Hodrick Prescott (1980) - Linear low pass filter
Baxter-King (1997) - Band Pass Filter
Beveridge-Nelson (1981) – Model based approach which
produces a stochastic trend
Unobserved Components Model – Uses a structural
model framework to models trend, cycle, and irregular
explicitly
5
Frequency versus Model Based
Frequency based filters extract a signal within a predefined
range. They implicitly apply a particular model to the data.
Hodrick and Prescott (1997) recommended a smoothing
parameter of 1600 based on an empirical investigation of US
quarterly GDP data.
Ravn (2002) recommended for annual data reducing the
parameter by a factor of four (approx 6.25)
Model based filters fit the model directly to the data. They
extract the signal from the estimated model.
Using the UCM to estimate the smoothing parameter  28.66
6
Low Pass Annual Filters
MFPt  Trendt  Cyclet  t
^
Cyclet
^
Cyclet  MFPt  Trend t
 MFPt  LowPass ( MFPt )
 1  LowPass  MFPt
 HighPass * MFPt
7
Frequency Response Function
11-Term Henderson
Hodrick-Prescott (6.25)
Hodrick-Prescott (28.66)
8
Frequency Response Function
11-Term Henderson
Hodrick-Prescott (6.25)
Hodrick-Prescott (28.66)
0.5
14.5 9.95 6.94
9
Market MFP Cycle Conparison
ABS Declared
Productivity Peaks
1994
10
Results
All methods (with the exception of the Beveridge-Nelson)
gave reasonably similar results
UCM verifies that the cycle component derived from HP
filter is not spurious.
UCM derived parameter is larger than theoretical value
Baxtor King requires more data at the end point
Results support the Hodrick-Prescott method.
More commonly used in practice internationally
Less likely to produce spurious cycles
11
Revision of Peaks
Looked at the amount of revision required using different methods
12
Revision of Peaks
13
Revision of Peaks
14
Revision of Troughs
15
Revision of Troughs
16
Key Findings From Study
There is strong evidence of cyclic behaviour in the MFP series
The Henderson filter suppresses more power in low cycles (> 8
years) and amplifies cycles from 4 to 6 years
May produce spurious cycles
Results support updating methodology to Hodrick Prescott filter
Smoothing parameter, theoretical (6.25) or derived (28.66)?
Impact of change will affect the 1994 peak, but should not
affect any other peaks
17
Comparing Peaks in Market MFP and
Industry MFP
Manufacturing and Construction have similar peaks to
Market MFP
Some industries (Electricity, Gas and Water, Wholesale) do
not show any cyclic behavior
 1999 Market MFP Peak:
PEAK
TROUGH
Construction
Mining
Accommodation, Cafes and
Restaurants
Electricity, Gas and Water
Communication
Transport and Storage
18
Further Work
Finalise smoothing parameter
Further analysis of Industry MFP
Analysis of growth between adjacent years relative to
average growth
Research capacity utilisation
19
Questions
20