The Anatomy of StartStop Growth

1
The Anatomy of Start-Stop Growth

• Ben Jones and Ben Olken
• October 11, 2005

2
Introduction

• Massive aid to poor countries at forefront of
  policy debate
• Sachs The End of Poverty
• G8 Debt forgiveness, foreign aid
• Idea
• Poor countries stuck in a poverty trap
• Massive influx of capital will get them out
• This paper examine the role of capital in actual
  experience of countries

3
Outline of talk

• First-order facts about within-country growth
• Identify structural breaks in growth
• Growth accounting around structural breaks to
  identify role of capital
• Potential sources of productivity gains
• Implications for growth models
• Conclusion

4
Related literature

• Decade-to-decade correlation in growth only 0.3
• Easterly et. al 1993
• Locating single structural break
• Pritchett 2000, Ben-David and Papell 1998
• Regressions on determinants of accelerations /
  decelerations
• Hausmann et al. 2005, Jerzmanowski 2005

5
Facts about growth
Best and Worst 10 Year Average Growth Rates
6
Facts about growth
7
Facts about growth
Growth spurts are not pure recovery
8
New growth

• These accelerations represent new growth, not
  just recoveries
• 80 of episodes show income expansions of 25 or
  more
• 50 of episodes show income expansions of 50 or
  more
• Only 6 do not exceed prior peak
• Similar results even over 15 year time periods
• Suggestions that understanding within-country
  growth may be paramount

9
Outline of talk

• First-order facts about within-country growth
• Identify structural breaks in growth
• Growth accounting around structural breaks
• Potential sources of productivity gains
• Implications for growth models (especially
  poverty traps)
• Conclusion

10
Identifying structural breaks

• Bai and Perron (1998, 2001)
• Conditional on given number of breaks, find break
  dates that maximize R2
• Propose tests to determine correct number of
  breaks based on additional R2 explained
• We conduct Monte Carlo simulations to estimate
  size / power
• Data All countries in Penn World Tables
  (1950-2000)
• We classify breaks into
• Upbreaks (accelerations)
• Downbreaks (decelerations)

11
Monte Carlos for Bai and Perron Method

• Generate synthetic growth processes
• 40 years of data
• Autocorrelation parameter of 0.1
• Introduce structural trend shifts of 0.5, 1, and
  2 times the standard deviation
• Results
• Appropriate size. BP method with 10 size
  produces false positives in 11 of cases
• Detects major events. Breaks of size 2s detected
  91 of time breaks of size 0.5s detected only
  24 of time

12
Some examples of breaks
13
Where are the breaks?
14
Outline of talk

• First-order facts about within-country growth
• Identify structural breaks in growth
• Growth accounting around structural breaks
• Potential sources of productivity gains
• Implications for growth models (especially
  poverty traps)
• Conclusion

15
Growth Accounting

• Standard aggregate production function
• Taking logs and differentiating
• Factors paid marginal products, output exhausted
  in factor payments, and allow for utilization of
  factors to vary
• Conduct this accounting exercise around
  structural breaks
• Short-run (5 years before vs. 5 years after
  break)
• Medium-run (entire growth regime before and after
  break)

16
Computing Capital Stocks

• Assume capital share a of 1/3
• Compute capital stock using perpetual inventory
  method
• Assume depreciation rate, d, is 7
• Calculate initial capital stock assuming initial
  growth rate of investment in time series prevails
  in pre-period
• Compute human capital from schooling
• Take Mincerian return, r, as 10
• Barro and Lee data (available every 5 years)

17
Growth Accounting Results
18
Assessing Utilization

• Labor utilization
• Data on labor force participation from ILO
• Capital utilization
• Use electricity consumption data from IEA
• Estimate relationship between electricity and
  capital using 1995 cross-section of countries
• R2 in regression is 90
• Estimate ? is 1.04 with .04 standard error.
• Suggests that electricity is linearly related to
  capital stock.
• Then
• Two uses of electricity data
• Can use electricity data in tandem with imputed
  capital growth to calculate capital utilization.
• Or use electricity growth rate to capture both
  utilization and accumulation i.e. avoid relying
  on capital imputation entirely

19
Growth Accounting Results
20
Statistical asymmetry tests
21
Summary of results

• Very little role for capital accumulation
• Negligible for accelerations
• Larger, but still relatively small, for
  decelerations
• Similar results if use electricity data only,
  rather than relying on national accounts
• Asymmetries are statistically and economically
  significant

22
Robustness

• Alternative method of choosing break dates
• Recoveries vs. new growth
• Lagged investment effects
• Assumptions about
• Depreciation rates
• Returns to scale

23
Alternative break dates

• Results robust to using Bai-Perron with different
  thresholds
• Alternative methods of selecting break dates
• Example Largest absolute changes in growth
• For each country, select two dates
• Acceleration date year where 10-year growth
  rate after minus 10-year growth rate before is
  largest
• Deceleration date year where 10-year growth
  rate after minus 10-year growth rate before is
  smallest
• Repeat growth accounting around
• All accelerations / decelerations
• Cases where change is greater than 5 percentage
  points

24
Largest accelerations / decelerations
25
Other Robustness Checks

• Growth accelerations as growth recoveries
• Restrict accelerations to cases where prior
  growth gt 0
• Try Hausmann et al. criteria for an acceleration
• Lagged investment effects
• Electricity consumption is a flow measure
• Collapses show immediate reaction in electricity
• Short-run similar to medium-run analysis
• Elasticities in aggregate production function
• Even AK model doesnt rescue investment for
  accelerations

26
Outline of talk

• First-order facts about within-country growth
• Identify structural breaks in growth
• Growth accounting around structural breaks
• Potential sources of productivity gains
• Implications for growth models (especially
  poverty traps)
• Conclusion

27
Sources of TFP changes

• Broadly speaking, TFP improvements could come
  from two areas
• Reallocation of resources (across or within
  sectors)
• Technology improvements within sectors
• Trade can lead to improvements in TFP
• Many mechanisms, including Ricardian
  reallocation, scale economies, knowledge
  spillovers, competition-induced
  efficiency/innovation
• Agriculture to manufacturing transition also
  associated with increased TFP
• Kuznets 1953, Matsuyama 1992, etc.
• Agricultural labor share averages 81 in poorest
  10 countries, only 7 in richest 10 countries

28
Movement in the Trade Share
29
Sources of TFP changes

• Taking Frankel and Romer (1999) estimates for
  impact of trade on growth, 25 expansion in trade
  implies 50-75 expansion in per-capita income
• Note that no major changes in terms of trade

30
Movement in the Manufacturing Labor Share
31
Sources of TFP changes
32
Other contemporaneous events

• What other important observables are changing
  around these breaks?
• Examine three types of variables
• Monetary policy
• Conflict (war)
• Institutions
• Descriptive exercise, not statements about
  causation

33
Other contemporaneous events
34
Movement in Inflation Rate
35
Changes in Internal Conflict
36
Other contemporaneous events

• Summary
• Downbreaks coincident with increased inflation,
  beginning of wars
• Upbreaks not typically associated with
  inflationary deceleration or end of conflict
• No particular change in institutions here
  corruption, rule of law, or democracy level. But
  causative evidence elsewhere for role of leader
  changes (Jones Olken 2005)
• Asymmetric roles of inflation and, to a lesser
  extent, conflict reinforce idea that the path
  into and out of growth accelerations are not the
  same

37
Outline of talk

• First-order facts about within-country growth
• Identify structural breaks in growth
• Growth accounting around structural breaks to
  identify role of capital
• Potential sources of productivity gains
• Implications for growth models (especially
  poverty traps)
• Conclusion

38
Implications (1)

• Stagnation is not a good description of poverty
• Nearly all countries have experienced rapid rates
  of growth 90 of grew faster than US for 10
  years or more
• Nearly all countries have also experienced
  declines, so these growth episodes rarely enough
  to escape trap
• While poverty traps may exist probabilistically,
  poor countries are capable of growth
• Within-country variance is first-order
• Average difference between best and worst 10-year
  growth rates within countries is 7 per annum for
  poorest 90 of countries

39
Implications (2)

• Short-run growth driven by TFP
• Even in neoclassical model, growth spurts can be
  driven by capital, even if steady-state growth
  due to TFP
• Starting in steady-state, consider change in
  investment / saving rate s. Large increase in
  growth due to capital accumulation possible if
  change s large
• Yet even in short run, we see its all TFP
• What would it take for capital to be the
  explanation?
• Recall
• For capital to explain accelerations, would
  require a 4.8
• If poverty trap, aggregate production function
  would need to be extremely convex

40
Implications (3)

• Transitions are asymmetric
• Suggestive evidence that very different factors
  involved in accelerations and collapses
• Accelerations trade
• Decelerations monetary instability, conflict
• Asymmetries in data are not found in most models
• May need different models for starting growth and
  sustaining growth

41
Final Example 1 Venezuela
Finland
Venezuela

• Per-capita income in Finland is 3.7 times larger
  than in Venezuela by 2000

42
Final Example 2 China

• Per-capita income in China is 2.7 times larger
  in 2000 than it would have been under pre-1978
  trend

43
Conclusion

• If LR growth is the summation of a few medium run
  experiences, then transitions between states seem
  of first-order importance
• Evidence from existing accelerations does point
  to capital influx as solution without assumptions
  of extraordinary elasticities
• Suggests focus on improving efficient allocation
  of resources, including classic stories based in
  trade and the transition to manufacturing

44
(No Transcript)
45
Unemployment?

• Labor force participation may not capture swings
  in unemployment
• But back-of-the-envelope calculation suggests
  its too small to matter
• Consider reduction in unemployment from 25 to
  10
• Would yield a one-time increase in income of
  maximum 15 (less since labor-share is only 2/3)
• By contrast, mean expansion in output is 5.2 for
  16.7 years in medium run, or increase in income
  of 240 over entire period

46
A poverty trap
kt1
kt