WEBVTT

NOTE The Phillips Curve as an Artifact of Austrian Business Cycle Theory

1
00:00:00.000 --> 00:00:05.960
When we estimate a Phillips curve, the relationship between inflation and unemployment, we expect

2
00:00:05.960 --> 00:00:11.040
to find a negative coefficient on inflation, but instead we often find that this coefficient

3
00:00:11.040 --> 00:00:13.160
is significant and positive.

4
00:00:13.160 --> 00:00:19.760
Furthermore, if we estimate a simple OLS autoregressive distributed lag specification of the kind proposed

5
00:00:19.760 --> 00:00:26.520
by William Niskanen in his 2002 Cato Journal article, we find that the coefficient on current

6
00:00:26.520 --> 00:00:31.400
Inflation is negative, consistent with the original Phillips curve and its Keynesian

7
00:00:31.400 --> 00:00:38.320
interpretation, but coefficients on lagged inflation are positive and significantly larger.

8
00:00:38.320 --> 00:00:44.600
This result can only be explained by Austrian business cycle theory. Inflation lowers unemployment

9
00:00:44.600 --> 00:00:49.960
in the short run while raising it and by a greater amount in the long run. The lowered

10
00:00:49.960 --> 00:00:55.340
unemployment initially brought about by inflation, the traditional Phillips curve, is the unsustainable

11
00:00:55.340 --> 00:00:59.000
Model Expansion Phase of the Austrian Business Cycle.

12
00:00:59.000 --> 00:01:05.020
The higher unemployment which follows is the process of malinvestment liquidation which

13
00:01:05.020 --> 00:01:08.700
cannot be explained by any other model of the business cycle.

14
00:01:08.700 --> 00:01:12.820
Niskanen's results were criticized to the effect that he should have estimated vector

15
00:01:12.820 --> 00:01:16.980
error correction models.

16
00:01:16.980 --> 00:01:22.840
We start by examining the consumer price index in a simple OLS regression where it is explained

17
00:01:22.840 --> 00:01:28.880
by Various Monetary Aggregates. These have very high R squares. Because all the data

18
00:01:28.880 --> 00:01:34.040
here are non-stationary, the estimates are super consistent, though residual tests suggest

19
00:01:34.040 --> 00:01:39.800
the monetary data should be put in logarithms. Recent empirical findings that monetary aggregates

20
00:01:39.800 --> 00:01:45.120
don't actually contribute to forecasting future consumer price index, combined with the fact

21
00:01:45.120 --> 00:01:50.380
that they do contribute here to explaining the current CPI, suggest that prices merely

22
00:01:50.380 --> 00:01:55.340
rise contemporaneously with the money supply.

23
00:01:55.340 --> 00:02:00.500
These residuals suggest structural changes with a marked increase in short-term volatility

24
00:02:00.500 --> 00:02:02.820
after 1995.

25
00:02:02.820 --> 00:02:07.140
In general, to overcome this problem, regressions and error correction models were estimated

26
00:02:07.140 --> 00:02:12.100
over truncated samples.

27
00:02:12.100 --> 00:02:17.820
Estimating a Phillips curve vector error correction model for 1948-2010, in the cointegrating

28
00:02:17.820 --> 00:02:22.420
In the following equation, the coefficient on inflation is positive and significant,

29
00:02:22.420 --> 00:02:27.260
meaning that higher inflation means higher unemployment, and the intercept suggests a

30
00:02:27.260 --> 00:02:30.940
natural rate of unemployment of about 4%.

31
00:02:30.940 --> 00:02:36.140
This result is robust to changing the lag specification in the disequilibrium adjustment

32
00:02:36.140 --> 00:02:42.900
process, truncating the sample range and adding the three-month Treasury bill interest rate.

33
00:02:42.900 --> 00:02:46.660
It could be questioned whether an error correction model is appropriate, however, because all

34
00:02:46.660 --> 00:02:56.200
All the data are stationary, that is, they don't have a long-term trend.

35
00:02:56.200 --> 00:03:01.780
The disequilibrium adjustment process, assuming 1 to 24 months of lag changes in unemployment

36
00:03:01.780 --> 00:03:08.540
and inflation, shows that every 1% in inflation raises unemployment by just under 1% after

37
00:03:08.540 --> 00:03:16.780
for years and this permanent increase in unemployment starts after about 14 months.

38
00:03:16.780 --> 00:03:23.060
Truncating the disequilibrium adjustment process to 6 to 18 months results in a slightly smaller

39
00:03:23.060 --> 00:03:28.680
total increase in unemployment with an earlier onset, but this seems to be an artifact of

40
00:03:28.680 --> 00:03:34.300
removing part of the actual process of adjustment.

41
00:03:34.300 --> 00:03:37.460
Adding the short-term interest rate gives the same result.

42
00:03:37.460 --> 00:03:42.600
Note that over the first two years, higher interest rates, looking now at the lower of

43
00:03:42.600 --> 00:03:48.840
these two graphs, lower unemployment slightly.

44
00:03:48.840 --> 00:03:52.980
This seems to be because the monetary expansion which results in inflation is implemented

45
00:03:52.980 --> 00:03:56.980
through targeting lower interest rates.

46
00:03:56.980 --> 00:04:04.500
This outcome is also robust to truncating the disequilibrium adjustment process.

47
00:04:04.500 --> 00:04:13.140
Now those were all over 1948 to 2010, over the period of 2003 to 2010, every 1.57% increase

48
00:04:13.140 --> 00:04:20.340
in inflation has resulted in a 4% increase in unemployment, an even less favorable trade-off.

49
00:04:20.340 --> 00:04:26.760
The period of onset of increased unemployment is also shortened.

50
00:04:26.760 --> 00:04:36.680
This is hardly changed by truncating the disequilibrium adjustment process.

51
00:04:36.680 --> 00:04:40.940
Next we examine a vector error correction model of two non-stationary processes, which

52
00:04:40.940 --> 00:04:45.920
are the kind of data that vector error correction models are really intended for.

53
00:04:45.920 --> 00:04:49.140
And these are civilian unemployment and the consumer price index.

54
00:04:49.140 --> 00:04:57.480
The two underlying data sets are variables that the unemployment rate and the inflation

55
00:04:57.480 --> 00:04:59.120
rate depend on.

56
00:04:59.120 --> 00:05:03.320
The cointegrating equation here gives us a positive and significant coefficient on the

57
00:05:03.320 --> 00:05:09.860
consumer price index, suggesting that a rising consumer price index accompanies higher employment

58
00:05:09.860 --> 00:05:11.400
in equilibrium.

59
00:05:11.400 --> 00:05:17.060
Now for the first error correction model, lower unemployment was good.

60
00:05:17.060 --> 00:05:46.500
Over 1948 to 2010, the disequilibrium adjustment process indicates a loss of 160,000 jobs for

61
00:05:46.500 --> 00:05:49.780
for every one point increase in the consumer price index.

62
00:05:49.780 --> 00:05:54.940
And here, as opposed to the earlier vector error correction model, the fact that it goes

63
00:05:54.940 --> 00:06:06.900
down indicates that when you raise the consumer price index, you lose jobs.

64
00:06:06.900 --> 00:06:12.460
Residual plots suggest structural changes in the CPI process, which is the one on the

65
00:06:12.460 --> 00:06:18.060
in the bottom, but not in the employment process, which is probably the more important one,

66
00:06:18.060 --> 00:06:24.460
which is the one on the top, because that's what we care about explaining.

67
00:06:24.460 --> 00:06:33.060
Over the 1982 to 2010 period, every one point increase in the CPI costs over 200,000 jobs,

68
00:06:33.060 --> 00:06:34.560
and it gets worse.

69
00:06:34.560 --> 00:06:39.580
Over the 1990 to 2010 period, every one point increase in the consumer price index costs

70
00:06:39.580 --> 00:06:45.740
There's over 480,000 jobs, almost half a million, but things aren't all.

71
00:06:45.740 --> 00:06:48.540
There is a silver lining.

72
00:06:48.540 --> 00:06:53.620
Over the 2003 to 2010 period, the most recent experience, every one point increase in the

73
00:06:53.620 --> 00:06:58.380
consumer price index costs a mere 409,000 jobs.

74
00:06:58.380 --> 00:07:03.500
Is this just an artifact of a combination of fixed weight CPI recently understating

75
00:07:03.500 --> 00:07:09.540
price increases and some actual price deflation which occurred during the recession and was

76
00:07:09.540 --> 00:07:12.900
is fairly unique to this period in the data set,

77
00:07:12.900 --> 00:07:17.460
or is the Keynesian resurgence actually improving the economy?

78
00:07:17.460 --> 00:07:18.140
No, it's not.

79
00:07:22.420 --> 00:07:25.020
Employment and MZM, money of zero maturity,

80
00:07:25.020 --> 00:07:27.720
are non-stationary variables for which vector error

81
00:07:27.720 --> 00:07:30.420
correction model is also appropriate.

82
00:07:30.420 --> 00:07:32.860
Increases in MZM lower employment

83
00:07:32.860 --> 00:07:36.380
in the cointegrating equation.

84
00:07:36.380 --> 00:07:39.020
This is the opposite of what was found with the consumer price

85
00:07:39.020 --> 00:07:40.060
Index.

86
00:07:40.060 --> 00:07:42.100
The coefficient is negative and significant

87
00:07:42.100 --> 00:07:46.500
and is robust to shortening the sample range.

88
00:07:46.500 --> 00:07:50.220
In the disequilibrium adjustment process for 1959 to 2010,

89
00:07:50.220 --> 00:07:54.960
every $1 billion increase in MZM wiped out 2,500 jobs

90
00:07:54.960 --> 00:07:58.260
after four years.

91
00:07:58.260 --> 00:08:00.660
Residual plots suggest structural instability

92
00:08:00.660 --> 00:08:03.900
in the MZM equation, but not in the employment equation.

93
00:08:03.900 --> 00:08:05.460
And generally, the interpretation

94
00:08:05.460 --> 00:08:15.580
One would be that MZM is more exogenous than employment, which depends on it.

95
00:08:15.580 --> 00:08:21.420
Since 1982, now this is with a truncated sample, every $1 billion increase in MZM has destroyed

96
00:08:21.420 --> 00:08:26.700
4,125 jobs.

97
00:08:26.700 --> 00:08:33.120
Since 1990, every $1 billion increase in MZM has destroyed nearly 10,000 jobs, and things

98
00:08:33.120 --> 00:08:35.640
These do get a little better.

99
00:08:35.640 --> 00:08:47.360
Since 2003, every $1 billion increase in MZM has only cost us 6,350 jobs.

100
00:08:47.360 --> 00:08:51.480
Another attractive vector error correction model to examine is in employment, the monetary

101
00:08:51.480 --> 00:08:54.300
base and the Austrian money supply.

102
00:08:54.300 --> 00:08:59.640
The monetary base is not statistically significant in the cointegrating equation, but the Austrian

103
00:08:59.640 --> 00:09:04.800
And money supply is always negative and significant.

104
00:09:04.800 --> 00:09:11.760
Over the 2003 to 2010 period, in the disequilibrium adjustment process, every $1 billion increase

105
00:09:11.760 --> 00:09:15.880
in the Austrian money supply wipes out approximately 5,000 jobs.

106
00:09:15.880 --> 00:09:21.000
And that's very consistent with what we found with MZM.

107
00:09:21.000 --> 00:09:26.160
These residuals don't suggest structural breaks, but there's less volatility in the 2009 to

108
00:09:26.160 --> 00:09:27.240
2010 period.

109
00:09:27.240 --> 00:09:56.240
This relationship between employment, unemployment, output, income, and general economic activity, on the one hand, and the money supply prices and inflation, on the other hand, is dramatically different from what has traditionally been supposed.

110
00:09:56.240 --> 00:10:05.240
The conventional view of the Phillips curve that there is a trade-off between inflation and unemployment is true at times, but is strictly a short-run phenomenon.

111
00:10:05.240 --> 00:10:20.240
Sometimes this relationship is reversed even in the short-run. It seems to be always a negative relationship in the long-run, the long-run meaning six to 48 months or longer, and a six-month long-run is not very long.

112
00:10:20.240 --> 00:10:24.880
That inflation decreases unemployment in the short run is consistent with the

113
00:10:24.880 --> 00:10:29.560
unsustainable expansion phase of Austrian business cycle theory. The much

114
00:10:29.560 --> 00:10:34.400
stronger and more robust finding that inflation increases unemployment and by

115
00:10:34.400 --> 00:10:38.680
much larger amounts over the longer run can only be explained by Austrian

116
00:10:38.680 --> 00:10:43.760
business cycle theory. This is an important result which strongly disconfirms all

117
00:10:43.760 --> 00:10:48.160
other models of the business cycle and this result is highly robust from a

118
00:10:48.160 --> 00:10:50.160
from a statistical perspective.