A FACTOR MODEL FOR WORLD TRADE GROWTH Elena Rusticelli and Stphanie - - PowerPoint PPT Presentation
A FACTOR MODEL FOR WORLD TRADE GROWTH Elena Rusticelli and Stphanie Guichard OECD Economics Department 6 th Colloquium on Modern Tools for Business Cycle Analysis Luxembourg 26-29 September 2010 Outline of the presentation 1. World trade
6th Colloquium on Modern Tools for Business Cycle Analysis Luxembourg 26-29 September 2010
The world trade growth is traditionally forecasted using a bottom-up approach where import and export volumes are forecasted on a country basis and the forecast for world trade is simply the aggregation of country-specific forecasts (OECD, IMF, WTO, World Bank). Short-term forecasting methods, corresponding to the direct approach, are normally considered as a useful benchmark which can point to possible up-downside risks to the current projections. Some studies (e.g. Burgert and Dées 2008) show the superior performance of direct forecasting methods where global factors play a fundamental role to explain world trade beside the traditional country-specific determinants.
12 13 14 15 16 2006 2007 2008 2009 2010 Trillions of 2005 US dollars Current world trade series EO87 projections (bottom-up approach) Bridge models projection (direct approach)
The main OECD tool for short-term forecasting of world trade growth - the bridge equation model - has been complemented with a new dynamic factor model which allows to extend the dataset
country level or different levels of aggregation without the risk of multicollinearity , losses of degrees of freedom and the increase in the computational burden
are available only with a shorter history
final forecasts, as well as their lagging or leading properties.
A large unbalanced dataset of 35 monthly indicators of different nature - hard, soft and financial indicators. Different levels of aggregation – global and country level, aggregate or disaggregate components. Stationarity achieved by means of monthly growth rates for all hard indicators ( except the Baltic Dry index) and the world share prices. Among survey indicators only the world stock level index has been transformed with first order differences.
Starting date Publication lags Source
ECONOMIC ACTIVITY World industrial production index 1991 2 CPB USA industrial production index 1991 2 CPB Japan industrial production index 1991 2 CPB Euro area industrial production index 1991 2 CPB Advanced economies industrial production index 1991 2 CPB Emerging economies industrial production index 1991 2 CPB Asia industrial production index 1991
2
CPB Latin Ameria industrial production index 1991 2 CPB Central and Eastern Europe industrial production index 1991 2 CPB Africa and Middle East industrial production index 1991 2 CPB Largest countries industrial production index 1990 2
OECD calculations
OECD retail sales 2000 3
OECD
World steel production 1980 1
IISI
SHIPPING AND FREIGHT ACTIVITY Baltic dry index 1985 1
The Baltic Exchange
Harpex shipping index 1996 1
Harper Petersen & Co.
International air traffic 1996 2
IATA
GLOBAL TECHNOLOGY CYCLE Tech pulse index 1971 1
CSIP
World semiconductor billings 1976 2
SIA
TRANSPORT COSTS Brent oil prices 1957 1
UK Dept. of Energy
EXPORT ORDERS G7 export orders 1962 1
OECD calculations
World export orders 1998 1
ISM
PURCHASING MANAGERS'INDEX Global PMI index 1998 1
ISM
PMI stock level index 1998 1
ISM
OECD + BRICS CLI 1960 2
OECD
World stock market prices index 1973 1
Datastream
US high yield spread 1984 1
OECD calculations
US loan officer survey (quarterly) 1990 1
FED
Monthly Indicators
HARD INDICATORS SOFT INDICATORS FINANCIAL INDICATORS
max lag
SIC value Ranking max lag
SIC value Ranking World industrial production (CPB) 0.91
1 0.80
1 World export orders 4 0.91
2 2 0.69
3 Largest countries industrial production 0.86
3 0.62
6 Global PMI index 2 0.84
4 2 0.69
2 Air freight volume 0.80
5 0.66
5 OECD+BRICS CLI 1 0.79
6 1 0.67
4 G7 export orders 2 0.75
7 1 0.49
8 US high yield spread 4 0.74
8 0.39
14 World stock market price 1 0.69
9
0.49
9 Baltic Dry Index 2 0.65
10 0.33
18 OECD retail sales 2 0.65
11 2 0.44
11 World steel production 1 0.61
12 0.34
16 Semi computers billings 0.57
13 0.46
10 PMI stock level index 0.54
14 1 0.51
7 US loan officer survey 2 0.53
15 0.41
13 US tech pulse index 0.52
16 0.43
12 Oil prices 0.52
17 0.37
15 Harpex index 0.42
18 0.34
17 Whole sample Sample ending in 2008 Q2 Max lag is based on the Schwarz criterion value, but ranking were not affected by changing the lag selection criteria
max lag
SIC value Ranking max lag
SIC value Ranking OECD+BRICS CLI 4 0.81
1 2 0.58
1 World stock market price 1 0.61
2 1 0.49
2 Baltic Dry Index 2 0.61
3 1 0.33
17 US high yield spread 3 0.60
4 2 0.41
7 Global PMI index 2 0.59
5 2 0.46
3 World export orders 2 0.58
6 2 0.44
5 Air freight volume 1 0.57
7 1 0.41
6 World industrial production (CPB) 1 0.56
8 1 0.38
10 OECD retail sales 2 0.55
9 2 0.39
9 G7 export orders 2 0.54
10 2 0.40
8 US loan officer survey 2 0.49
11 1 0.37
11 PMI stock level index 1 0.46
12 1 0.44
4 Largest countries industrial production 2 0.46
13 1 0.35
14 World steel production 1 0.45
14 1 0.33
18 Oil prices 2 0.44
15 1 0.33
16 US tech pulse index 2 0.42
16 1 0.35
13 Harpex index 1 0.36
17 1 0.33
15 Semi computers billings 1 0.35
18 1 0.37
12 Max lag is based on the Schwarz criterion value, but ranking were not affected by changing the lag selection criteria Whole sample Sample ending in 2008 Q2
World industrial production index World exports orders Largest countries industrial production index Global PMI index
.00 .02 .04 .06
.00 .02 .04 90 92 94 96 98 00 02 04 06 08
.00 .02 .04 .06
.00 .02 .04 90 92 94 96 98 00 02 04 06 08 32 36 40 44 48 52 56 60
.00 .02 .04 90 92 94 96 98 00 02 04 06 08 36 40 44 48 52 56 60 64
.00 .02 .04 90 92 94 96 98 00 02 04 06 08
OECD+BRICS CLI World stock market price index Baltic Dry index US high yield spread
2 4 6 8 10 12 14 16 18
.00 .02 .04 .06 90 92 94 96 98 00 02 04 06 08 2,000 4,000 6,000 8,000 10,000 12,000 14,000
.00 .02 .04 90 92 94 96 98 00 02 04 06 08
.0 .1 .2 .3
.00 .02 .04 90 92 94 96 98 00 02 04 06 08 97 98 99 100 101 102 103 104
.00 .02 .04 90 92 94 96 98 00 02 04 06 08
and tech pulse index)
Bayesian conditional VAR to forecast monthly indicators Quarterly bridge equation, i.e. ADL(p,q) to forecast world trade growth
=
+
=0
− +
where
=
1,
, … ,
,
is a ( × 1) vector of monthly indicators and Bs a ( × ) matrix of coefficients. where
and
represents the quarterly world trade growth rate and all aggregated monthly indicators expressed in growth rates, except for export orders.
=
+
− =1
+
,
, −
=0 =1
+
The dynamic factor model is To combine the monthly factor model with the quarterly world trade growth, the latent monthly world trade growth variable is
=
+
~ℕ0, Σ
=
=1
− +
=
~ℕ0,
=
′
+
~ℕ(0,
2)
where the ( × 1) vector of monthly indicators
=
1,
, . . ,
,
=
1,
, . ,
,
=
1,
, . . ,
,
driven by q variable-specific shocks
=
1,
, . . ,
,
, with ≤ .
= 1
3 (
+
−1 +
−2)
In the a monthly state space representation are estimated
= Λ 1
+
−
′
1 − 1 3 1
+1
+1
_1 =
A
1
Ξ
+1
+
+1
+ℎ| =
( , ℎ)
− −1 =0
where
=
,
and the dataset downloaded at time t is equal to ℤ
=
=0 , with
( , ℎ) =
( + , ℎ) with p>0 and t large enough, hence
( , ℎ) =
(ℎ).
as the bridge model
as the bridge model plus 9 more IP indicators on a country /macro- regional level
dynamic factor model DFM2 plus 10 more indicators on a world level
Forecasting error measures over the period 2003 Q1 - 2007 Q4
QUARTERS AR BM DFM1 DFM2 DFM3 MAE Previous 0.84 0.64 0.46 0.39 0.34 Current 0.71 0.80 0.73 0.77 0.59 One-quarter-ahead 0.96 0.83 0.67 0.75 0.69 Two-quarters-ahead 1.03 0.98 0.78 0.80 0.77 Average 0.89 0.81 0.66 0.68 0.60 MAPE Previous 0.71 0.72 0.50 0.35 0.35 Current 0.60 0.91 0.85 0.77 0.74 One-quarter-ahead 1.00 0.80 0.93 0.89 0.83 Two-quarters-ahead 1.50 0.95 0.88 0.90 0.85 Average 0.95 0.85 0.79 0.73 0.69 RMSE Previous 1.03 0.74 0.51 0.46 0.40 Current 0.92 0.93 0.87 0.86 0.76 One-quarter-ahead 1.25 0.98 0.81 0.88 0.85 Two-quarters-ahead 1.32 1.11 0.91 0.93 0.92 Average 1.13 0.94 0.78 0.78 0.73 FORECASTING MODELS
World trade growth rates forecasts over the period 2010 Q2 - 2011 Q1, with OECD world trade series of goods and services and monthly indicators published by the end of July 2010.
1 2 3 4 5 Previous Current One-quarter-ahead Two-quarters-ahead DFM3 DFM2 DFM1 BM
Cumulative forecast weights of different indicators over the sequence of 4 forecasts for the new dynamic factor model
1 2 3
Previous Current One-quarter-ahead Two-quarters-ahead PMIs Industrial production Technology indicators US high yield spread Shipping rates and freight World share prices Other
Dynamic factor models can be a useful tool to forecast short- term world trade growth The forecasting accuracy of these models is higher than the bridge equation models. They enable to include relevant monthly indicators with a more recent starting point. Different contributions of aggregate and disaggregate components and country versus world level data can be assessed. The dataset will be extended to include country breakdowns for more monthly indicators: PMIs, retail sales, etc. Inclusion of quarterly indicators as US loan officers survey.