1. Is Private Investment in Advanced Economies Too Low?*
Title page
Is Private Investment in Advanced Economies Too Low?*
Seok Gil Park
Research Department
* Preliminary version, please do not cite or circulate.

2. Summary Question / Methodology
Question: Is investment in AEs in line with output dynamics?
Has investment behavior changed after the crisis?
Models: Output accelerator. Checking with uncertainty, Tobin’s Q, etc.
Countries: France, Germany, Italy, Spain, Japan, United States
Findings
Investment explained by output dynamics, once allowing for a change in coefficients after the crisis.
Firms’ target capital stock (to output ratio) increased during ’s, but returned to previous level after GFC

3. Weak Investment in AEs: Disappointing Recovery
Contributions to Forecast Errors:
(percentage-point difference between actual and predicted growth 1/)
“Financial markets have been optimistic, … However, this has not translated into a pickup in investment, which—particularly in advanced economies—has remained subdued.” (WEO 2014 October)
Growth forecast errors mostly from unexpectedly low investment
Sources: IMF, World Economic Outlook; and staff calculations.
1/ Average of one-year-ahead forecast errors for the years from 2011 to / Crisis economies defined as in Laeven and Valencia (2013).

4. Methodology and Data

5. Accelerator Model An old idea:
Optimal (target) capital stock as a linear function of output level
Then investment should respond to changes in output level (Accelerator).
Txt slide

6. Model Variations 1) Accelerator + Uncertainty
Real I/K ratio, as a function of output accelerator and uncertainty (Unc)
2) Augmented Tobin’s Q model
Controls for firms’ financial status (alternative Tobin’s Q, operating surplus), user cost of capital (Cost), and capacity utilization ratio (Util)
3) All-inclusive model
Txt slide

7. Data Definition / Robustness Check
Investment: Real, gross fixed capital formation for equipments
Capital stock: Real net stock of equipments, extrapolated with implied depreciation rate
Uncertainty: Standard deviation of output accelerator
Alternative Q: Equity to Asset ratio of nonfinancial corporations
User cost of capital: (Real) Bank lending rate, adjusted for relative price of investment goods (with GDP deflators)
<Robustness Check>
Using output level as denominator instead of capital stock
Txt slide

8. Regression Results

9. Investment Became Less Sensitive to Accelerators
Structural Break: In overall assessments, impact of output accelerator on investment decreased after GFC.
Robust: for most of countries (U.S. an exception) and most of specifications (See tables in Appendix)
Coefficients for accelerator1/
p-value for
Adj. R squared
All sample
Pre-crisis2/
Post-crisis3/
Chow test
w/o SB
w/ SB
Pooled sample
1.202***
1.300***
0.414***
0.000
0.935
France
2.874***
2.334***
0.535***
0.715
0.853
Germany
0.939***
0.953**
0.513***
0.561
0.265
0.240
Italy
1.569***
1.509***
0.030
0.566
0.650
Japan
1.467***
1.977***
0.866***
0.438
0.554
Spain
1.600***
1.850***
0.856***
0.885
0.913
United States
0.601
1.363
1.957***
0.001
-0.032
0.055
Source: Haver Analytics; IMF staff calculation. Model was estimated with Newey-West type standard error.
1/ Sum of coefficients for lagged accelerators. 2/ 1990Q1 – 2008Q4, 3/ 2009Q1 – 2014 Q2
3/ *** and ** denote statistical significance at the 1% and 5% level, respectively.

10. Recent Trends Explained with Structural Break
Real Private Investment: Equipment
(In percent of GDP; Fitted values are from the all-inclusive model without structural break)
(In percent of GDP; Fitted values are from the all-inclusive model with structural break)
Source: Haver Analytics; IMF staff calculation. Note/ Standard errors of predictions were calculated from simple OLS.

11. Puzzle, If Any, Maybe from Ignoring Structural Break
Residuals from Regression
(In percent of GDP, Average )
Source: Haver Analytics; IMF staff calculation

12. Then Why Structural Break? A Peek from Time-Varying β
Rolling Window Regression Coefficient for Accelerator
(40-Quarters window, equipment investment, accelerator model without uncertainty)
Source: Haver Analytics; IMF staff calculation

13. Time-Varying β: Using Y as denominator
Rolling Window Regression Coefficient for Accelerator
(40-Quarters window, equipment investment, accelerator model without uncertainty, output level as denominator)
Source: Haver Analytics; IMF staff calculation

14. Time-Varying β: Shorter Rolling Window
Rolling Window Regression Coefficient for Accelerator
(20-Quarters window, equipment investment, regressed only on average accelerator)
Source: Haver Analytics; IMF staff calculation

15. Interpretations
Ratio of target capital stock to output fluctuated, partly reflecting longer-frequency business cycle.
For example, IT boom in 1990’s seems to have raised the ratio (accelerator coefficient) notably in US, but the coefficient returned to previous levels.
Returning to mediocre normal?
Longer-history data suggest that the coefficient had not been significantly high before 1990’s.
Secular stagnation?
Some downward trend in the coefficient since 2000’s
Txt slide

16. Appendix: Regression Table

17. Regression Coefficients, Equipment Investment
France
Pre
Post
Output accelerator1/
2.334
0.535
2.651
0.826
2.842
0.616
2.248
0.355
2.222
0.565
1.834
0.800
1.732
0.701
(0.000)
(0.001)
(0.003)
0.853
Uncertainty
-0.931
-0.287
1.791
0.366
-0.074
0.098
(0.488)
(0.179)
(0.060)
(0.929)
(0.045)
0.875
Tobins Q ratio
0.100
1.179
-1.806
-1.246
2.745
-0.577
0.071
-1.154
(0.951)
(0.392)
(0.084)
(0.005)
(0.958)
(0.013)
0.672
0.876
Surplus
0.316
0.425
0.066
0.280
-0.119
0.294
-0.142
0.104
(0.099)
(0.641)
(0.388)
(0.122)
0.670
0.852
User cost of capital
0.108
-0.134
0.090
0.026
0.311
0.001
0.158
-0.056
(0.095)
(0.070)
(0.390)
(0.921)
(0.023)
0.688
0.910
Utilization ratio
17.249
4.578
4.605
-1.970
13.940
2.547
9.864
-1.055
(0.440)
(0.175)
(0.407)
0.777
0.829
0.889
0.908
Germany
Output accelerator
0.953
0.513
1.429
0.511
0.845
0.459
1.292
0.373
1.241
0.510
-0.611
0.292
0.509
0.220
(0.018)
(0.008)
(0.022)
(0.036)
(0.083)
(0.534)
(0.009)
0.240
(0.561)
-1.247
-0.204
-2.047
-0.002
-0.476
-0.048
(0.056)
(0.053)
(0.963)
(0.648)
(0.030)
0.223
0.542
4.191
2.136
4.310
0.478
1.071
0.428
1.673
(0.002)
(0.185)
(0.451)
(0.180)
(0.172)
(0.155)
0.417
0.506
0.168
0.178
0.210
0.070
0.114
0.043
0.138
0.103
(0.043)
(0.012)
(0.047)
0.372
(0.025)
0.603
0.133
0.044
-0.006
0.067
0.061
0.142
0.075
(0.203)
(0.619)
(0.686)
(0.685)
(0.399)
(0.093)
0.094
(0.466)
0.175
(0.660)
13.316
3.673
20.683
1.518
7.913
2.692
4.708
(0.120)
(0.459)
(0.787)
0.560
0.537
0.622
0.660
Source: Haver Analytics; IMF staff calculation. 1/ Sum of coefficients for lagged accelerators
2/ Figures within parenthesis in second rows are p-values for coefficients. Figures in italics in third rows (left columns) are OLS adjusted R-squared. Figures in italics within parenthesis in third rows (right columns) are p-values from Chow test.

18. Regression Coefficients, Equipment Investment
Italy
Pre
Post
Output accelerator1/
1.509
0.030
1.512
0.871
1.179
0.377
0.907
0.675
1.601
0.630
0.799
0.821
1.155
0.989
(0.007)
(0.926)
(0.009)
(0.000)
(0.384)
(0.008)
(0.002)
(0.004)
(0.048)
0.650
Uncertainty
0.070
0.594
0.158
0.999
1.146
1.199
(0.907)
(0.001)
(0.798)
0.536
0.681
Tobins Q ratio
0.398
-0.106
0.677
0.779
0.523
-0.827
0.151
0.363
(0.112)
(0.739)
(0.431)
(0.234)
(0.073)
(0.029)
0.781
0.832
Surplus
0.020
-0.005
0.057
-0.543
-0.112
-0.010
0.037
0.254
(0.793)
(0.976)
(0.511)
(0.010)
(0.177)
(0.951)
(0.494)
0.769
0.801
User cost of capital
-0.057
-0.117
-0.052
-0.320
-0.019
-0.167
-0.067
-0.140
(0.062)
(0.550)
(0.013)
(0.005)
0.628
0.780
Utilization ratio
8.876
-0.329
6.268
-7.361
5.282
0.524
2.312
1.567
(0.819)
(0.053)
(0.036)
(0.682)
(0.022)
(0.044)
0.687
0.704
0.811
0.939
Spain
Output accelerator
1.850
0.856
1.851
0.496
1.767
0.900
1.855
0.861
2.260
0.886
1.714
0.696
1.690
0.646
0.913
-0.174
-0.726
0.007
-0.314
-0.589
-0.401
(0.567)
(0.927)
0.529
0.912
1.761
1.602
0.493
1.110
2.648
2.231
2.036
(0.201)
(0.216)
(0.054)
0.595
0.916
0.014
-0.066
-0.068
-0.065
0.026
(0.773)
(0.207)
(0.651)
(0.854)
(0.122)
(0.049)
0.516
0.911
0.035
0.029
0.059
-0.033
0.013
0.109
0.034
0.032
(0.591)
(0.732)
(0.023)
(0.498)
(0.731)
(0.003)
(0.046)
0.741
0.963
19.203
6.077
2.678
1.454
12.557
5.293
4.836
-2.327
(0.231)
(0.071)
0.752
0.928
0.988
Source: Haver Analytics; IMF staff calculation. 1/ Sum of coefficients for lagged accelerators
2/ Figures within parenthesis in second rows are p-values for coefficients. Figures in italics in third rows (left columns) are OLS adjusted R-squared. Figures in italics within parenthesis in third rows (right columns) are p-values from Chow test.

19. Regression Coefficients, Equipment Investment
Japan
Pre
Post
Output accelerator1/
1.977
0.866
2.490
0.425
1.948
0.814
1.049
0.543
2.100
0.794
0.484
0.579
0.551
0.302
(0.000)
(0.291)
(0.001)
(0.323)
(0.006)
0.552
Uncertainty
1.059
-0.883
-0.820
-0.702
-2.115
-0.855
(0.204)
(0.222)
(0.021)
0.142
(0.040)
0.605
Tobins Q ratio
1.258
0.512
1.142
0.388
-0.067
-1.013
0.494
-0.206
(0.167)
(0.038)
(0.819)
(0.030)
(0.081)
(0.225)
0.355
(0.072)
0.632
Surplus
0.463
0.152
0.361
0.124
0.254
0.188
-0.077
-0.185
(0.022)
(0.161)
(0.220)
(0.702)
(0.013)
0.606
User cost of capital
0.008
-0.042
-0.034
-0.045
-0.062
0.007
-0.043
(0.891)
(0.052)
(0.347)
(0.014)
-0.021
(0.649)
0.581
Utilization ratio
3.613
0.789
3.073
0.822
1.168
0.372
2.229
1.399
(0.153)
(0.395)
(0.681)
(0.029)
0.712
0.748
0.732
0.783
United States
Output accelerator
1.363
1.957
0.679
0.968
-0.276
1.590
1.183
1.024
2.565
2.300
3.110
-0.250
1.302
-0.059
(0.169)
(0.523)
(0.827)
(0.140)
(0.023)
(0.008)
(0.074)
(0.768)
0.055
-3.595
-2.237
-3.083
-1.262
-0.215
0.231
(0.059)
(0.181)
(0.002)
(0.606)
(0.349)
0.262
(0.062)
0.165
1.776
2.953
2.030
0.894
1.944
0.380
1.256
-0.069
(0.009)
(0.025)
(0.343)
(0.746)
0.281
0.187
0.350
0.382
0.346
0.201
0.133
0.341
0.024
(0.037)
(0.186)
0.577
0.591
-0.055
-0.666
-0.221
0.258
0.160
-0.093
0.196
0.037
(0.555)
(0.024)
(0.018)
(0.203)
(0.686)
0.062
0.199
-6.134
15.280
18.741
7.928
18.368
(0.108)
(0.004)
(0.073)
0.131
0.333
0.882
0.913
Source: Haver Analytics; IMF staff calculation. 1/ Sum of coefficients for lagged accelerators
2/ Figures within parenthesis in second rows are p-values for coefficients. Figures in italics in third rows (left columns) are OLS adjusted R-squared. Figures in italics within parenthesis in third rows (right columns) are p-values from Chow test.