BIS Working Papers

No 812
Steady-state growth

by Emanuel Kohlscheen and Jouchi Nakajima

Monetary and Economic Department

September 2019

JEL classification: C11; C15; E30; O40

Keywords: economic growth; financial conditions;

inflation; monetary policy; potential output; time-
varying parameter VAR; trend growth
BIS Working Papers are written by members of the Monetary and Economic
Department of the Bank for International Settlements, and from time to time by other
economists, and are published by the Bank. The papers are on subjects of topical
interest and are technical in character. The views expressed in them are those of their
authors and not necessarily the views of the BIS.

This publication is available on the BIS website (www.bis.org).

© Bank for International Settlements 2019. All rights reserved. Brief excerpts may be
reproduced or translated provided the source is stated.

ISSN 1020-0959 (print)

ISSN 1682-7678 (online)
 Steady-State Growth∗
E. Kohlscheen and J. Nakajima†

Abstract
We compute steady-state economic growth - defined as the rate
of growth that the economy would converge to in the absence of new
shocks. This rate can be computed in real-time by means of a parsimo-
nious time-varying parameter (TVP) VAR model. Our procedure of-
fers a relatively agnostic estimation of benchmark equilibrium growth
rates. Estimates show that the steady-state GDP growth rate in the
case of the United States declined from just above 3% per year in the
1990s to 2.4% at present. Results for other six advanced economies
and the euro area indicate that the steady-state growth rate, which
is consistent with stable inflation and financial conditions, has been
relatively stable since 2010 in most cases in spite of a recent slowdown
in actual GDP growth rates.
JEL codes: C11; C15; E30; O40.
Keywords: economic growth; financial conditions; inflation; mon-
etary policy; potential output; time-varying parameter VAR; trend
growth.

∗
The views expressed in this paper are those of the authors and do not necessarily
reflect those of the Bank for International Settlements or the Bank of Japan. We thank
Claudio Borio, Angelo Fasolo, Andrew Filardo, Wagner Piazza Gaglianone, Marco Lom-
bardi, Dubravko Mihaljek, Benoit Mojon, Hyun Song Shin, Christian Upper and seminar
participants at the BIS for useful comments and suggestions.
†
Monetary and Economic Department, Bank for International Settlements. Central-
bahnplatz 2, 4051 Basel, Switzerland.

1
1 Introduction

What is the medium to long-term GDP growth rate of an economy? And

against which benchmark should current growth rates be assessed at any

point in time? Policy makers are continuously confronted with these ques-

tions. The answer to them is crucial for the appropriate calibration of mone-

tary and fiscal policy. This paper presents a relatively agnostic estimation of

steady-state growth rates, based on a time-varying parameter (TVP) struc-

tural VAR model.

Our definition of steady-state growth is simply the growth rate to which

an economy would converge to in the absence of any new shocks. An economy

is modelled by means of a parsimonious and agnostic TVP-VAR model. The

steady-state growth rates can be easily recomputed in real time after each

new observation, making it a useful reference in practice.

In comparison with standard HP-filter techniques, an important advan-

tage of the TVP-VAR methodology is that it is not haunted by end-point

distortions (see Hamilton (2018)). At the same time, our benchmark growth

rate relies on less a priori theoretical restrictions than the well-known model

of Laubach and Williams (2003) for the natural rate of interest. Further-

more, the allowance for variation of the parameters over time enables the

model to continuously adapt to changes in policy or private sector behavior

1
(Primiceri (2005)). The TVP-VAR model features stochastic volatility of
1
This implies that the model is more robust to potential mutations in the Phillips

2
innovations, so that changes in the properties of the shocks are well captured.

We apply our method to assess the steady-state growth rates for seven

OECD countries and the euro area. The estimates show that the steady-

state growth in the case of the United States declined from just above 3%

per year in the 1990s to 2.4% at present. Results for other countries indicate

that the steady-state growth rate - which is consistent with stable inflation

and normal financial conditions - has been relatively stable in most cases in

recent years despite a recent slowdown of the actual GDP growth rates. The

methodology can be easily replicated for other countries.

Relation to the literature. Several studies have tried to estimate

equilibrium growth rates in the past using a variety of approaches. Methods

differ on how much they rely on theoretical models. Our approach, which is

strictly empirical, uses the TVP-VAR model developed by Primiceri (2005)

and del Negro and Primiceri (2015). Lubik and Matthes (2015) have al-

ready employed this methodology to compute natural interest rates in the

United States. Earlier, Laubach and Williams (2003) constructed a small-

scale macroeconomic model and used a Kalman filtering method to compute

the natural rate of interest and trend growth rates. Their crucial assumption

is that the output gap is determined by its own lags and the ex ante real funds

rate gap. Holston, Laubach and Williams (2017) discuss how estimates of

that model may vary according to model specification. Our work is inspired
curve.

3
by this evolving literature.

The main advantage of our analysis is that the model parameters are

not assumed to be fixed, but are allowed to vary over time as new data

points come in. In this sense, the approach is far less prescriptive, and the

model is simpler and more flexible. It is also more appropriate if behaviors,

policies and responses change over time. On the whole, the contribution of

this paper is to provide a benchmark steady-state growth rate that can be

easily estimated in real-time.

Outline. The paper proceeds as follows. Section 2 explains the econo-

metric specification. Section 3 presents estimated steady-state growth rates

for several major OECD countries and the robustness of these estimates. We

show the differences between real time and full sample estimates and the

sensitivity to alternative prior specifications. The last section offers some

concluding remarks.

2 Computing Steady-State Growth Rates

We estimate the standard TVP-VAR model

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
∆ ∆−1 ∆−2
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ ∆ ⎦ =  + 1 ⎣ ∆−1 ⎦ + 2 ⎣ ∆−2 ⎦ +   (1)
 −1 −2

where ∆ stands for the log change in output, ∆ for the log change in

the core CPI, and  represents an indicator of monetary policy or financial

4
 2
conditions;  is the time-varying intercept; and 1 and 2 are matri-

ces of coefficients, which are time varying.   represents the vector of het-

eroskedastic disturbances, which follows the multivariate normal distribution

as   ∼ (0  ), where  is the time-varying covariance matrix.  is de-

composed to the variance and covariance components as  = −1 0 −10

    ,

where  is the diagonal matrix with the diagonal elements corresponding to

the variance of structural shocks, and  is a lower-triangular matrix with

diagonal elements equal to one. We denote as   the vector of elements in

 for  = 1 2;   as the vector of diagonal elements of  ; and  as the

vector of free parameters in  .

As in Primiceri (2005), the time varying parameters follow the first-order

autoregressive process

 = −1 +  

 1 =  1−1 + 1 

 2 =  2−1 + 2 

 = −1 +  

log   = log  −1 +   

where  ,  ,  and   are innovations. In other words, the coefficient vec-

tors are modelled as random walks and the standard deviations are geometric
3
random walks - making it a stochastic volatility model.
2
The selection of variables is such that it allows the model to have one implicit IS
curve, a Phillips curve and a Taylor rule.
3
A more detailed description of the model specification and estimation methodology

5
 In what follows, we use the Goldman Sachs financial conditions index or

the 5-year government bond yield as  . The latter is preferred to the short-

term policy rate because it better reflects the overall stance of monetary

policy during the recent period of policy rates at or close to the zero lower

bound.

The steady-state or equilibrium values can be derived from (1). More

specifically, after rearranging (1), the steady-state values have to satisfy

⎡ ⎤
∆
⎢ ⎥ b1 − 
b2 )−1b
⎣ ∆ ⎦ = ( −   

In what follows, we focus on the benchmark growth rate ∆ .

3 Estimation

3.1 Benchmark Results

To obtain the steady-state growth rates, the TVP-VAR models were esti-

mated for seven OECD countries and the euro area. Estimates were based

on quarterly data. Our benchmark steady state growth rate estimation uses

the Goldman Sachs financial conditions indices (FCIs). Later, as a robust-

ness check, we show estimates when the 5-year government bond yield is used

as the variable . Due to the availability of the FCI, the sample period starts
for the TVP-VAR model can be found in Primiceri (2005) and Nakajima (2011).

6
in 1990Q3 and ends in 2018Q4. For Canada, Sweden and Switzerland, the

starting point of sample period varies according to data availability. For the

euro area, time series were taken from the AWM database with the sample

period spanning from 1990Q3 to 2017Q4.

Following Primiceri (2005) and Nakajima (2011), the prior distribution

for the variances of innovations  ,  ,  and   is set as an inverse gamma

distribution with the hyperparameters calibrated by a pre-sample period.

The number of lags of the VAR is two based on the Bayesian Information

Criterion (BIC).

[TABLE 1 about here]

Table 1 reports steady-state growth rates for the 7 largest advanced

economies and the euro area just before the GFC (2006Q4) and at the end of

our sample (i.e. 2018Q4). Equilibrium growth in the United States declined

from an estimated 2.8% per year in 2006 to 2.4% in 2018. Similarly, in the

United Kingdom it declined from 2.3% per year to 2.0%. Estimates were less

affected by the crisis in the cases of the euro area and Japan, where they

stand at, respectively, 2.0% and 1.2%. Canada and Switzerland also exhibit

a slight decline in the steady-state growth rate between 2006 and 2018.

[GRAPH 1 about here]

Graph 1 plots the evolution of the estimated steady-state growth rates

over time. The solid lines show the mean estimate and the shadow areas

7
Table 1. Steady-state growth rates.

%
2006Q4 2018Q4
United States 2.8 2.4
Euro area 1.8 2.0
Japan 1.1 1.2
United Kingdom 2.3 2.0
Canada 2.5 2.3
Australia 2.7 2.8
Switzerland 2.0 1.9
Sweden 2.2 2.6
Note: 1. For the euro area and Switzerland, 10-year yield is used. 2.
For the euro area, the latest estimate refers to 2017Q4.
Figure 1. Steady-state growth rate (VAR with FCI).

% United States % Euro area

4.5 4.0

4.0 3.5

3.5 3.0

3.0 2.5

2.5 2.0

2.0 1.5

1.5 1.0

1.0 0.5
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Japan % United Kingdom

3.5 4.5

3.0 4.0

2.5 3.5

2.0 3.0

1.5 2.5

1.0 2.0

0.5 1.5

0.0 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Canada % Australia
4.5 4.5

4.0 4.0

3.5 3.5

3.0 3.0

2.5 2.5

2.0 2.0

1.5 1.5

1.0 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Switzerland % Sweden
4.0 4.5

3.5 4.0

3.0 3.5

2.5 3.0

2.0 2.5

1.5 2.0

1.0 1.5
0.5 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

Note: For the euro area (up to Q4 2017) and Switzerland, 10-year yield is used. Shadow area indicates 68% confidence intervals.
indicate the 68% confidence intervals. Estimates show that steady-state GDP

growth rate in the case of the United States declined from just above 3% per

year in the 1990s to the current 2.4%. For the euro area, one noteworthy

aspect is that the confidence band is substantially wider than elsewhere.

Large uncertainty in this case hardly comes as a surprise given the shorter

time series and the regional heterogeneity (e.g. in labor markets).

Graph A1 in the Appendix plots the evolution when the 5-year govern-

ment bond yield is used as  instead. By and large, the patterns are found
4
to be similar. Further, Graph A2 shows the evolution of the stochastic

volatility of growth shocks. As can be seen in the plots, the financial crisis
5
led to a spike in the volatility of growth innovations in most cases.

3.2 Comparison with other Approaches

It may also be useful to compare the estimates obtained from our methodol-

ogy to existing potential growth estimates. The OECD’s production function

based estimates of potential growth for the median country in 2006 are ex-

actly equal to the median delivered by the TVP-VAR approach for that
6
year (i.e. 2.3% per year). However, while the median potential output
4
We also find similar patterns and levels when we use the shadow interest rates of Wu
and Xia (2016) for the United States, United Kingdom and the euro area. The confidence
bands are however wider under this choice during the period that interest rates are close
to zero.
5
The exception is Australia, where core inflation increased most after the crisis, to
above 5%. The model considered this unsustainable. The trend growth rate recovered
when core inflation dropped.
6
Based on data from the OECD Economic Outlook no.105, May 2019.

8
growth has declined 69 basis points according to the OECD, our econometric

methodology leads to a reduction of only 10 basis points by 2018 for the me-

dian country. The methodology seems to reflect that higher growth rates are

compatible with stable inflation and financial conditions. We find that the

bulk of the reduction in trend growth rates in most countries predates the

GFC (as can clearly be seen in Graph 1). In the case of the United States,

however, we find that the trend growth rate diminished by a further 40 basis
7
points since 2006.

The current rate of 2.4% for the United States is somewhat higher than

the 2.0% estimated by Holston, Laubach and Williams (2017). Also our

estimate for the euro area trend growth is above theirs. In contrast, the

estimate for the United Kingdom is virtually identical as our estimate is

2.0% while theirs is 1.9%.

3.3 Sensitivity to Priors

The estimated steady-state growth rate could be affected by the prior speci-

fication for the innovation variance of the time-varying intercept  and the

lagged coefficients  1 and  2 . In the baseline model, we use the inverse

gamma distribution, IG( = 25, = 00025). To check for robustness of the

estimates, we estimate the model with alternative prior hyperparameters: (1)

IG(10, 0.0025), (2) IG(50, 0.0025), (3) IG(25, 0.001), and (4) IG(25, 0.005).
7
Put differently, the model does not interpret the financial crisis as a very substantial
change in the transmission mechanism, but it sees it as a large negative output shock.

9
 Table 2 reports the posterior means and 68% confidence intervals of the

steady-state growth rate for the Unitest States in 2006Q4 and 2018Q4. The

estimates turn out to be only marginally different, which implies that the

estimation methodology is quite robust. We also examined the sensitivity

for the other countries and found that the posterior estimates of steady-state

growth rate do not change much with alternative prior specifications.

[TABLE 2 about here]

3.4 Real-Time

Table 2 reports the difference between real-time and full-sample estimates

in each case. In the case of the G-3 economies (the United States, the euro

area and Japan) corrections for steady-state growth rates are generally found

to be modest. For 2016Q4 and 2017Q4, for instance, they are never above

15 basis points. Larger corrections for 2016Q4 are found in the case of the

United Kingdom.

[TABLE 3 about here]

On average, two-years ago real-time steady state growth rates are ad-

justed by 19 basis points, while one-year ago estimates are adjusted by 12

basis points. All in all, this assessment suggests that a reasonably reliable

estimator of equilibrium growth rates is obtained in real-time in most cases.

10
Table 2. Prior sensitivity: US steady-state growth rate estimates

%
2006Q4 2018Q4
Lower Mean Upper Lower Mean Upper
Baseline 2.36 2.82 3.24 2.02 2.44 2.85
Prior 1 2.24 2.84 3.37 1.88 2.32 2.77
Prior 2 2.39 2.78 3.16 2.12 2.50 2.87
Prior 3 2.41 2.79 3.16 2.14 2.53 2.91
Prior 4 2.26 2.84 3.37 1.92 2.36 2.79
Note: "Lower" and "Upper" correspond to 68% confidence intervals.

Table 3. Difference in steady-state growth rate

estimates between real time and full sample

%
2016Q4 2017Q4
United States -0.13 0.10
Euro area -0.11 -0.04
Japan 0.05 0.10
United Kingdom 0.58 0.06
Canada -0.06 0.34
Australia 0.44 -0.01
Switzerland 0.03 0.29
Sweden -0.11 0.01
Note: Real-time estimate less the full-sample (up to 2018Q4). For
the euro area, 2015Q4 and 2016Q4.
4 Concluding Remarks

We presented a new benchmark growth rate. The approach to obtain the

steady-state growth rate is purely data driven, and does not rely on fixed

relations between economic variables. The model is adaptative, as changing

coefficients are updated after each new data point. The TVP-VAR based

methodology is not subject to end-of-sample distortions and offers flexibility

as it captures for instance changing Phillips curves. An additional advantage

is that the steady-state growth rate can easily be computed in real time.

The estimates show that steady-state GDP growth rate for the United

States declined from just above 3% per year in the 1990s to 2.4% at present.

In six other advanced economies and the euro area the steady-state growth

rate, which is consistent with stable inflation and normal financial conditions,

has been relatively stable after the crisis in most cases, despite a recent

slowdown in actual GDP growth rates.

11
 References

del Negro, M. and G. Primiceri (2015) "Time varying structural vector

autoregressions and monetary policy: A corrigendum", Review of Economic

Studies, vol. 82, pp. 1342-1345.

Hamilton, J.D. (2018) Why you should never use the Hodrick-Prescott

filter. Review of Economics and Statistics, vol. 100, pp. 831-843.

Holston, K., T. Laubach, and J. C. Williams (2017) "Measuring the nat-

ural rate of interest: International trends and determinants", Journal of

International Economics, vol. 108, pp. 59-75.

Laubach, T. and J. C. Williams (2003) "Measuring the natural rate of in-

terest", Review of Economics and Statistics, vol. 85, pp. 1063-1070.

Lubik, T. A. and C. Matthes (2015) "Calculating the natural rate of inter-

est: A comparison of two alternative approaches", Economic Brief, no. EB15-

10, Federal Reserve Bank of Richmond.

Nakajima, J. (2011) "Time-varying parameter VAR model with stochastic

volatility: An overview of methodology and empirical applications", Mone-

tary and Economic Studies, vol. 29, pp. 107-142.

Primiceri, G. (2005) "Time varying structural autoregressions and mone-

tary policy", Review of Economic Studies, vol. 72, pp. 821-852.

Wu, J. C. and D. Xia (2016) "Measuring the macroeconomic impact of

monetary policy at the zero lower bound", Journal of Money, Credit, and

Banking, vol. 48(2-3), pp. 253-291.

12
Figure A1. Steady-state growth rate (VAR with 5-year yields).

% United States % United Kingdom

4.5 4.5

4.0 4.0

3.5 3.5

3.0 3.0

2.5 2.5

2.0 2.0

1.5 1.5

1.0 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Canada % Australia
4.5 4.5

4.0 4.0

3.5 3.5

3.0 3.0

2.5 2.5

2.0 2.0

1.5 1.5

1.0 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

Note: Shadow area indicates 68% confidence intervals.

Figure A2. Stochastic volatility for the growth rate (VAR with FCI).

% United States % Euro area

5.0 5.0

4.0 4.0

3.0 3.0

2.0 2.0

1.0 1.0

0.0 0.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Japan % United Kingdom

10.0 6.0

8.0 5.0

4.0
6.0
3.0
4.0
2.0

2.0 1.0

0.0 0.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Canada % Australia
4.5 4.5
4.0
4.0
3.5
3.5
3.0
3.0
2.5
2.5
2.0
1.5 2.0

1.0 1.5

0.5 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

% Switzerland % Sweden
4.0 7.0

3.5
6.0
3.0
5.0
2.5
4.0
2.0
3.0
1.5

1.0 2.0

0.5 1.0
92 94 96 98 00 02 04 06 08 10 12 14 16 18 92 94 96 98 00 02 04 06 08 10 12 14 16 18

Note: For the euro area (up to Q4 2017) and Switzerland, 10-year yield is used. Shadow area indicates 68% confidence intervals.
Previous volumes in this series

811 Embedded supervision: how to build Raphael Auer

September 2019 regulation into blockchain finance

810 Spillovers of funding dry-ups Iñaki Aldasoro, Florian Balke,

September 2019 Andreas Barth and Egemen Eren

809 Inflation expectations anchoring: new insights Nikos Apokoritis, Gabriele Galati,
September 2019 from micro evidence of a survey at high- Richhild Moessner and Federica
frequency and of distributions Teppa

808 A disaster under-(re)insurance puzzle: Home Hiro Ito and Robert N McCauley
August 2019 bias in disaster risk-bearing

807 Bank intermediation activity in a low interest Michael Brei, Claudio Borio and
August 2019 rate environment Leonardo Gambacorta

806 Geographic spread of currency trading: the Yin Wong Cheung,

August 2019 renminbi and other EM currencies Robert N McCauley and Chang Shu

805 Exchange rate puzzles: evidence from rigidly Charles Engel and Feng Zhu
August 2019 fixed nominal exchange rate systems

804 (Un)conventional Policy and the Effective Fiorella De Fiore and Oreste Tristani
August 2019 Lower Bound

803 Determinants of credit growth and the bank José Bustamante, Walter Cuba and
July 2019 lending channel in Peru: a loan level analysis Rafael Nivin

802 A loan-level analysis of bank lending in Carlos Cantú, Roberto Lobato,

July 2019 Mexico Calixto López and Fabrizio López-
Gallo

801 The internationalization of domestic banks Paola Morales, Daniel Osorio and
July 2019 and the credit channel: an empirical Juan Sebastián Lemus-Esquivel
assessment

800 Banks’ business model and credit supply in Miguel Biron, Felipe Cordova and
July 2019 Chile: the role of a state-owned bank Antonio Lemus

799 Monetary policy surprises and employment: Rodrigo Barbone Gonzalez

July 2019 evidence from matched bank-firm loan data
on the bank lending-channel

798 How do bank-specific characteristics affect Carlos Cantú, Stijn Claessens and
July 2019 lending? New evidence based on credit Leonardo Gambacorta
registry data from Latin America

All volumes are available on our website www.bis.org.