Transcript Presentation

Assessing the predictive power of
measures of financial conditions for
macroeconomic variables
Kostas Tsatsaronis
Head of Financial Institutions
Bank for International Settlements
Bank of Greece,
4 February 2010
1
Real and financial sector interactions
Financial
sector
Real
sector
2
Real and financial sector interactions
 Take the “real” sector point of view
– How does the financial sector influence the
macroeconomic picture?
 Forecasting: better understand business cycle
 Modelling: stylised facts about interaction between
business and financial cycle
 Policy:
– Information content of financial variables
– The reaction function of monetary policy
3
Objective
Question: Can we summarise the links between financial
conditions and the macroeconomy in a single simple
measure?
Yardstick: How do measures of financial conditions fare as
forecasters of macroeconomic variables in the one-to-two
year horizon.
Variables: GDP Gap, Investment, inflation
Countries: United States, Germany, United Kingdom
4
Methodological approach
 Non-model driven econometrics
 Data intensive but not a predominately structural approach
– Establish stylised facts
 Examine different economies
5
Results
 Financial conditions factors have important information
content
 Financial conditions factors have independent information
content:
• Information is complementary to asset prices
 Financial conditions factors have more information content
for real variables than for inflation
 Financial conditions factors perform better at longer
horizons
6
Summarising financial conditions
 Distil common information from a large number of
variables into small number of factors
– Stock and Watson (2002)
 Focus exclusively on financial variables
 Use as many as possible
 Representing as broad an array of financial sector activity
as possible
 Keep the balance between prices and quantities
7
Summarising financial conditions
Statistical procedure creating latent factors (Principal
Components)
Int. rates + spreads
Asset prices
Credit
F1 , F2 , F3 , …
Performance of
financial institutions
--------------------------~ 40 variables
Focus: top-6 latent factors ~ 50% of total variance
8
Data
 Bank assets and liabilities & income statements
 Interest rates
 Exchange rates
 Equity market indicators
 Real estate indicators
 Flow of funds variables
 Balance of payments variables
 Other
9
Data handling
 Deal with stationarity
 Perform normalisation
 Quarterly interpolation of annual series
– Project annual series onto annualised factors
– Use mapping to interpolate into quarterly
• Flow and stock variables
• Level ad first differenced series
10
11
Forecasting
y t  k  Own lags  Inflation 
n
   lj Fl ,t  j  t  k
l 1,,6  j  0
Financial conditions
Specification: lag and factors selection to optimise BIC
(trade-off between goodness of fit and parsimony)
12
13
14
Results
 Financial conditions factors have information content
• Significant coefficients
• Output and investment: good
Inflation: not so good
 Overall forecasting performance quite good:
• R2 range 40-85%
• Not so sharp decline in longer horizon
 Small number of factors
• Explain 20% of variance
• Stable set across horizons
15
Horse race against asset prices
 Is the informational content of the financial
factors essentially the same as that of the yield
curve and equity prices?
 Horse race regression (encompassing)
y t  k  Asset prices  
l
  lj Fl ,t  j  t  k
j
16
Table 3
“Horse race” against selected asset prices: predicting the output gap
US
Germany
UK
k=4
k=8
k=4
k=8
k=4
k=8
R-sq adj
61%
42%
50%
44%
91%
75%
Excl. PCs
0.121
--
0.003
0.001
0.0003
0.0001
Excl.
Other
0.035
0.419
0.011
0.971
0.0000
0.0000
17
A Financial Conditions Index?
y t  k  Own lags  Inflation 
n
   lj Fl ,t  j  t  k
l 1,,6  j  0
Financial conditions
 The linear combination of the principal components
represents a relationship among financial variables that is
correlated forward with real variables:
• Positive values are good for the economy
• Negative values are harmful
18
A Financial Conditions Index?
 The weights of the original data are fairly
constant across different lags
• One could construct an FCI using only
contemporaneous values of the original
series and then take lags of this composite
series
19
Future work
 Expand the set of countries in the analysis
 Examine for threshold and asymmetric effects in the
relationship between financial and real variables
 How stable is the composition of the FCI?
– Out of sample performance
20