Transcript
Stages of Development and Evolving Constraints to Growth Xuehui Han and Shang-Jin Wei Asian Development Bank May 27, 2015 The views expressed in this document are those of the author and do not necessarily reflect the views and policies of the Asian Development Bank or its Board of Governors or the governments they represent.
1
Roadmap • Background, Motivation, and Overview • Is there a “Middle-Income Trap”? – Transition matrix and ergodic distribution
• What are the leading constraints to growth at different stages of development? – Conditional inference tree (and random forest)
• Conclusions 2
Is there a “Middle-Income Trap”?
Google scholar citations: 126,000 Google citations: 2,650,000
3
What is a trap? • Merriam-Webster Dictionary: a situation or a position from which it is difficult or impossible to escape • Oxford dictionary: a device or an enclosure … typically by allowing entry but not exit … • So a “middle-income trap” should be …?
4
• What has been said to be a middle-income trap? • Selected discussions on Middle-Income Traps: – Growth slowdown at $10,000–$11,000 and $15,000– $16,000 in 2005 PPP (Eichengreen et al. 2013); – A typical country in the middle-income group has a higher frequency of deviations from the growth path (Aiyar et al., 2013); – Might be a trap in relative term (Robertson and Ye, 2013). 5
Define the thresholds for the income groups Extremely Low Income
LowIncome
$1,096 GDP per capita (2005 ppp); $3 per day
WB definitions
LowerMiddleIncome
UpperMiddleIncome
HighIncome
$2,418; WB’s cut-off line
$5,500; median point between $2,418 and $15,220
$15,220; US’s income in 1960
$2,418
$10,276
$19,429
GDP per capita (2005 PPP)
6
11
Figure 1. Income Transition from 1960 to 2011
LUX NOR
9
TPE
KOR ROU
BWA GNQ
THA CHN TUN
8 7
LSO BFA MLI ETH
MOZ
BDI
NLDDNK AUS AUTBEL IRL SWE CAN FINDEU ISL GBR FRA ESP ITA NZL MLT GRC ISR PRT TTO BRB
ARG TURCHL MYS MEX GAB IRN PAN CRI MUS BRA DOMPER COL ZAF ECU
USACHE 45 Degree Line
URY VEN
CPVIDN
COD
6
HKG JPN CYP
JORNAM JAM LKA FJI PRY ZWE GTM BOL SYR IND MAR PHL HND MRT COG GHA PAK NGA ZMB TCD CMR BGD SEN CIV TZA GMB KEN BEN RWA UGA NPL SLV GIN TGO COM GNB MWI MDG CAF $2,418 (in 2005 PPP) NER
EGY
6
logIncome2011
10
SGP
Low Income
Middle Income
7
8
$15,220 High Income 9
10
logIncome1960
7
Decade-average transition matrix for 1960–2010 (in %) Extremely LowIncome EL 82 L 3 LM 0 UM 0 H 0 Ergodic distribution 0
LowIncome
LowerMiddleIncome
UpperMiddleIncome
HighIncome
18 72 3 0 0
0 25 68 0 0
0 0 29 70 0
0 0 0 30 100
0
0
0
100
8
• Formally, there is no unconditional “middle income trap.” • There is no unconditional “low income trap” either. • The only unconditional trap is a “high income trap”
9
# of decades needed for X percent of countries to be in higher income groups
X=50% X=90%
ELIC
LIC
LMIC
UMIC
4 14
3 12
3 8
2 7
10
• However, one might get a difference picture if one looks at incomes relative to the US
11
1.2
Relative Income Transition (in % of US income) 1960–2011 NOR
SGP
HKG
NLD AUT SWE DNK BEL CAN FIN DEU FRAISLGBR ITA
0.2
0.20
URY ZAF
0.4
PER
DOM
COL
0.15
BRB ECU
TUN
EGY
IDN CPV IND
LKA FJI
PRY BOL MAR
JOR
PAK
GTM
SYR
PHL GHA
NGA
ZWE
TCD CMR
BGD LSO TZA BEN GMB KEN UGA RWA SLV NPL BFA MLI MWI TGO COMGNB MOZ ETH MDG CAF NER BDI COD
0.6
0.05
0.8
NAM
HND
MRT COG
0.00
0.0
MUS
BRA
THA CHN
0.00
0.2
0.4
TTO ARG TURCHL ROU MYS MEX GAB PAN BWA IRN CRI GNQTHA BRA PER MUS COL ZMB CHN DOM TUN ECU NAM JOR JAM EGY LKA FJI IDN PRY GTM CPV BOL SYR IND MAR PHL HND MRT PAK GHA ZWE COG NGA TCD CMR BGD LSO SEN16% TZA KEN CIV NPL UGA RWA BEN GMB BFA SLV MLI TGO COM GNB GIN of US MOZ ETH MWI MDG CAF BDI NER COD
Enlarged Portion of Income below 16%
NZL GNQ
ISR
0.10
MLT GRC PRT
ESP
Income2011RelativeUS
CYP
0.05
0.8 0.6
KOR TPE
USA [$15,220(1960),$42,646(2011)]
AUS
IRL JPN
0.0
Income2011RelativeUS
1.0
CHE VEN
SEN
CIV GIN
16% of US
0.10
1.0
Income1960RelativeUS
0.15
0.20
1.2
Income1960RelativeUS
12
Decade-average transition matrix for 1960–2010 relative to US (in %) 16% and below
16%–36%
36%–75%
75% and above
16% and below
92
8
0
0
16%–36%
13
72
15
0
36%–75%
0
4
73
22
75% and above
0
2
19
79
23
13
31
33
Ergodic distribution
13
• A relative income trap is a bit silly, similar to saying that the human specie has an “infant age trap” because newborns can never catch up with their parents in age.
14
Moving beyond transition matrix analysis • The transition matrix analysis has an important assumption: • Every country in a given income group is similar to all other countries in the same group • But different countries in a group can have different fundamentals and policy choices, and these may matter for growth too. 15
What are the binding constraints to growth at different stages of development?
16
What is a Regression Tree? • Non-parametric Machine Learning method proposed by Breiman (1984); • Searches for all possible binary splitting points for each predictor; • Newly developed Conditional Inference Tree: – No structures imposed; – Can deal with many regressors; missing data; correlations between predictors; – Good at picking up the leading explanatory variables. 17
Indicators 1. Political Stability: Domestic conflicts indication ; 2. Macroeconomic stability: inflation, government debt share, and the number of crisis episodes (currency and banking);
3. Infrastructure: paved roads & railway, power generating capacity. 4. Human capital: the average years of schooling; 5. Openness: trade/GDP; FDI net inflow/GDP; 6. Good governance: political constraints indices 7. Initial Income level; 8. Demography: share of the 15-64 cohort in population; growth rate of the 15-64 cohort; 9. Growth rate of the leading economics; 10. Oil exporter indicator.
ADB Washington Others Consensus President Nakao’s Eight Key Actions
18
Tree Growing • • • •
p ≤ 0.05 for splitting; Minimum size for splitting: 20 obs.; Minimum size for ending nodes: 7 obs.; Using 10-year horizon for growth rate measurement; • Most explanatory variables are decade averages.
19
Low-income group
If unfavorable demographics, low inflation + good infrastructure
Unfavorable demographics + high inflation
If not resource Oil exporter rich, favorable + favorable demographics demographics + good 20 education
Three groups of countries • Progressive (annual growth>3%; blue circle); • Near-stagnant (≤3%), • Regressive (negative annual growth; red triangle);
21
Variable combinations Progressive group Oil exporter + favorable demographics If Not oil exporter, then favorable demographics + good education If unfavorable demographics, then low inflation + good infrastructure Regressive Unfavorable demographics + high inflation
Country examples
Growth rates
Indonesia (1970) Sudan (2000) India (1990) Sri Lanka (1980)
8.6% 5.2% 4.2% 4.7%
Cape Verde (1980) Malaysia (1960)
4.0% 2.0%
Madagascar (1980) Tanzania (1990) Tanzania (1980) Zambia (1990)
–0.7% –1.0% –4.5% –3.0% 22
What is a Forest? • Many trees (1000 trees in our forest); • For each tree, use 90% of the sample for tree growing; • Count the appearance of the predictors across all trees and take average; • Each tree is subject to different errors. By taking the average of all trees, we have more robust results than the regression tree. 23
Low-Income
Rank
(1) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Factors
(2) Population share (15–64) Paved Road (km/1,000 workers) FDI share Power Generating Capacity (gigawatts/1,000 workers) Log Initial Income Population growth (15–64) Oil Exporter or Not Years of Schooling Inflation Government Debt Share Global Growth Rate Log Conflicts Index Political Constraints Railway (km/1,000 workers) Trade Share No. of Crises
Difference of Frequencies of growth between positive Average nodes with higher Total contributions / split / low value than frequency negative value the splitting value contributions (in (decade growth %) rate) (3) (4) (5) (6) 1277 53.77% 2.54 93.89/6.11 1181 1.39 1.21 69.18/30.82 955 1.83% 0.79 65.03/34.97 839
0.12
1.54
78.67/21.33
832 828 740 736 617 603 571 507 480 294 282 278
6.86 0.2 0 3.21 12.83% 41.82% 2% 5.5 0.12 0.39 60.44% 3
-2.47 2.37 2.36 2.02 -2.21 -1.72 0.84 -1.12 1.62 0.29 -0.92 -1.39
9.62/90.38 85.75/14.25 78.65/21.35 89.95/10.05 15.24/84.76 5.14/94.86 71.98/28.02 30.97/69.03 78.96/21.04 53.06/46.94 20.21/79.79 24.82/75.18 24
Middle-income group
If unfavorable demographics + low debt + low initial income
Unfavorable demographics + high debt + low openness + low global growth
If unfavorable demographics + high debt+ high openness
Favorable demographics + low debt
25
Variable combinations Country examples Progressive Favorable demographics + low PRC (2000) debt Greece (1960) Portugal (1990) Japan (1970) If unfavorable demographics + high debt, then high openness to FDI If unfavorable demographics, then low debt + low initial income Regressive Unfavorable demographics + high debt + low openness + low global growth
Growth rates 8.1% 7.9% 4.5% 4.1%
Malaysia (1970) Bolivia (2000)
7.6% 3.5%
Turkey (1960) Chile (1960)
3.5% 3.4%
Honduras (1980) Sri Lanka (1970)
–0.7% –4.5%
26
Middle-Income Rank
Factors
(1) 1 2 3 4 5 6 7 8 9 10 11 12 13
(2) Population share (15–64) Government Debt Share No. of Crisis Log Initial Income FDI share Global Growth Rate Political Constraints Years of Schooling Inflation Population growth (15–64) Log Conflicts Index Oil Exporter or Not Trade Share Power Generating Capacity (gigawatts/1,000 workers) Paved Road (km/1,000 workers) Railway (km/1,000 workers)
14
15 16
Difference of growth Frequencies of between nodes with positive Total Average Split higher / low value than contributions / Frequency Value the splitting value negative (decade growth rate) contributions (in %) (3) (4) (5) (6) 2093 59.03% 2.38 96.42/3.58 1690 38.29% -2.01 2.9/97.1 1414 8 -2.02 4.03/95.97 1152 8.51 -2.37 7.12/92.88 1036 1.84% 2.25 91.89/8.11 974 2% 2.04 97.33/2.67 721 0.28 -0.15 51.87/48.13 606 6.41 0.22 45.87/54.13 603 13.98% -0.64 32.01/67.99 505 0.18 -0.98 22.57/77.43 488 5.75 0.58 59.84/40.16 458 0 1.92 89.96/10.04 422 62.75% -1.22 27.49/72.51 286
0.8
-0.46
41.96/58.04
256
4.32
1.09
75.39/24.61
246
0.69
1.19
68.7/31.3 27
Using Regression Tree Results to Predict Decade Transition Matrix • Classify countries into three groups: progressive (annual growth>3%; blue circle), near-stagnant (≤3%), and regressive (negative annual growth; red triangle); • Apply the “end node” predicted growth rates to each country’s initial income to get the predicted decade end income; • Construct the decade transition matrix. 28
Regression-Tree-Simulated Transition Matrix (in %) obs. in the obs. group / percentage total obs. Extremely Low-Income Countries Progressive 12/82 14.63 Near-stagnant 60/82 73.17 Regressive 10/82 12.20 Low-Income Countries Progressive 32/104 30.77 Near-stagnant 59/104 56.73 Regressive 13/104 12.50 Lower-Middle-Income Countries Progressive 43/86 50.00 Near-stagnant 19/86 22.09 Regressive 24/86 27.91 Upper-Middle-Income Countries Progressive 66/97 68.04 Near-stagnant 23/97 23.71 Regressive 8/97 8.25
Extremely LowIncome
# years # years for 50% for 90% moving moving up up
LowIncome
LowerMiddleIncome
UpperMiddleIncome
HighIncome
17 85 100
83 15 0
0 0 0
0 0 0
0 0 0
4 43 never
13 142 never
0 0 54
38 92 46
62 8 0
0 0 0
0 0 0
8 79 never
24 261 never
0 0 0
0 0 0
44 95 100
56 5 0
0 0 0
9 129 never
29 426 never
0 0 0
0 0 0
0 0 12
65 96 88
35 4 0
17 156 never
54 518 never 29
Conclusions (1) • We reject the unconditional notion of a “middleincome trap,” or a “low-income trap”. • In general, there is no trap for countries in a progressive group. • For countries in a regressive group (minority), the negative expected growth rate implies that they may do worse than being simply trapped in their current income status. • For countries in a near-stagnant group, because growth is low, they may look like they are being trapped in their current income status for a long time. 30
Conclusions (2) • For low-income countries, favorable demographics, macroeconomic stability, good education, and good infrastructure appear to be the most important variables separating fast growing and slow growing countries. • For middle-income countries, favorable demographics, macroeconomic stability, sound global economic growth, and openness to FDI appear to be the key discriminatory variables. 31
