Cours de Politique Economique

Document Sample
Cours de Politique Economique Powered By Docstoc
					    Comments on the Papers of session 1 A

By Gilles Dufrénot (University of Aix-Marseille and Banque de France)
Plan of the discussion
Le cours s’articule autour de trois points:
The questions, the methodology and the main results
 Still unsolved puzzles (!)

II.-SPECIFIC COMMENTS (methodological aspects)
 Paper by Bulligan
 Paper by Alvarez and Cabrero
 Paper by Ferrara and Vigna
Le cours s’articule autour de trois points:

Why the « agnostic »
 approach leaves us with
 unsolved puzzles
What are we interested in?
Le cours s’articule autour de trois points:
Three important goals:

    Assess the vulnerability of the economy to Housing markets
     (Hence the importance of finding cross correlations)

    Try to identify the main channels of transmission of the shocks
     from the housing sector to the economy (hence the importance of
     using several indicators of the housing activity)

    Prospective approach : find leads and lags dynamics to say whether
     housing variables can be chosen as advanced indicators of the GDP
How can we do that?
  Theoretical models : typically Dynamic Stochastic
  General Equilibrium models  growing literature
  (appropriate to study how productivity shocks,
  technological shocks, monetary shocks, etc… affect the
  agents’ decisions of investment in the housing sectors
  and the impact on the GDP

  Statistical models (« agnostic » about the main
  transmission channels)  only want to detect
  correlations, leads and lags dynamics, common factors
  in the series
The three papers adopt the second
approach (1)
  1.- Country-specific studies based on time

      The paper by Bulligan deals with the case of Italy and
       use a structural VAR framework to study the effects of
       monetary policy shocks  What’s new : the sign
       restriction approach in the long-run matrix to
       define the sequance of shocks

      The paper by Alvarez and Cabreo deals with the case
       of Spain  What’s new : butterwoth filters to study
       the cross –correlation between variables of the
       housing sector and macroeconomic variables
The three papers adopt the
second approach (2)
  1.- Country-specific studies based on time series

      The paper by Ferrara and Vigna deals with the case of France.

       What’s new : The second part of the paper (!) that
       discusses the long-run cross correlations betwwen the
       housing markets and the economic variables; this allows
       them to explain why France was less affected by the sharp
       decrease in the housing prices
The three papers adopt the
second approach (3)
  2.- All three papers yields similar conclusions

       The find strong correlations between the business cycle and
        the cycle of the housing activity (whichever variables are used
        to capture the dynamics in either sector or the other)

       They conclude that the activity in the housing sector can be
        considered as a leading indicator of the business cycles.
First unsolved puzzle (1)
 1.- What about the house price gap ?

      Which variables were at play during the last 10 years to explain
       the booming in the housing investment and the upward price

             Fundamental variables : growth of disposable income per –
              capita, long-term and short-term interest rates, inward
              immigration flows (Spain), credit growth, changes in equity
              prices, etc..

             Non-fundamental variables (% of increase in the housing
              prices , not explained by the fundamentals)
First unsolved puzzle (2)
First unsolved puzzle (3)
  If non-fundamentals are at play during the booms, then
  we can expect a sharp correction when the prices and
  activity in the housing sector decline.

  So, during upwards, when the house price are over-
  valued, we expect to find a weaker correlation with the
  GDP (because you have a bubble)
First unsolved puzzle (4)
  Now, when the bubbles burst, there are two indications:

  Firstly, they do so when some macroeconomic fundamentals
   begin to deteriorate (income, unemployment, credit
   conditions, etc….  In this case the GDP growth may be a
   leading indicator of the activity in the housing sector (!)

  Secondly, deteriorating macroeconomic conditions are the
   source of downward revisions in expectations  Stronger
   correlations betweeen the GDP and the activity in the
   housing sector
First unsolved puzzle (5)
 Implications for the time series-based approaches

  Selection bias if the period includes episodes of huge
  price decrease (may explain the strong positive
  correlations that are found)

  Concerning the lead and lags effects: the approaches do
  not handle the house price gap
Second unsolved puzzle (1)
  Spurious short-term cross-correlation ?

   Except in Spain and Ireland, the residential investment
   does not account for a large share of the economies.

   Ratio of housing construction in % of GDP : 6,5% for
    the advanced economies and over the past 3 decades :
   Accordingly, for the correlation between the housing
    sector and the GDP to be strong, there must be large
    corrections in the housing construction!!
Second unsolved puzzle (2)
  Two consequences for the time series models

   Again a problem of selection bias : the correlation found
   include periods of huge -downward - corrections in the

   Choice of the variables: even if we accept the idea that
   the housing sector activity leads the GDP, the variable of
   interest should be the investment rate in this sector (real
   residential investment to GDP)
Third unsolved puzzle
  The papers focus too much on the short-term, but the
   long-run correlation may also be important

   The papers argue that cyclical componenst of the
   housing variables affect the cyclical upturns and
   downturns of the GDP.

   However, for downturns in the GDP, it is known that
   they occur in the industrialized countries when the
   ratio of housing investment to GDP evolve below its
   historical trend !

   This implies that the trend components of the housing
   variables affect the cyclical components of the GDP
Le cours s’articule autour de trois points:

Paper by Guido Bulligan
Cyclical andtrend growth
  Business approach

   The old methodology by Burns and Mitchell has been
   updated by Harding and Pagan (2002), Journal of
   Monetary Economics  link between the turning points
   and the moments of the series + cycles are obtained as
   regards their contribution to volatility, trend growth,
   correlation and non-linear effects.

  Missing nonlinearities in the series + structural
   Use non-parametric filters such as polyspectra
    (bispectraum, trispectrum) + evolutionary sprectrum
VAR models (1)
  VAR analysis to study the implication of monetary
   policy shocks

   Isn’t there a problem of « multiscale »: housing markets
   are characterized by long cycles with a persitent
   dynamics, as compared with the other macroeconomic
   variables in the VAR? Is it possible to estimate the effects
   of a shock by considering a VECM?

   There is a similar study as yours done by Carlos Vargas-
   Silva (2009) for the US (forthcoming in the Journal of
   Macroeconomics), showing that
VAR models (2)
  VAR analysis to study the implication of monetary
   policy shocks

  1/ the magnitude of monetary shocks on the housing
    markets is very dependent on the selection of the
    horizon for which the restrictions hold in the VAR ;

  2/ as compared with classical choleschi decomposition ,,
    the impact of monetary policy on the housing market is
    much less certain with the sign restriction approach.

  Do you find similar things for Italy?
Le cours s’articule autour de trois points:

Paper by Luis Alvarez and
 Alberto Cabrero
   There is one filter that may overperform the ones
   described : wavelet for several reasons (simultaneous
   description of the high frequency and low-frequency

   Compared with the butterworth filter, you do not need
   to eliminate some « high » or « low » frequencies,
   because the filter is « multiscale »

   Compared with the Kernel regression, Wavelet
   decomposition is also non-parametric, but the analysis
   is done in the « frequency domain » which is more
   appropriate for the study of business cycles then time
   domain methodologies
Comparison with DSGE models (1)
   One problem : your empirical results sometimes
   contradict the theoretical findings.

   But, it does not mean that the DSGE models are wrong!
  The negative or positive response to shocks depends upon

  1/ the interval of variation of the parameters that serve to
    calivrate the models and
  2/ upon the nature of the technological, productivity,
    monetary shocks.

  How can you use your time-series based filters to see
   whether your findings do indeed contradict the
   conclusions of the models?
Comparison with DSGE models (2)
  Three steps (Monte Carlo) :

   1/ obtain simulated series from the DSGE models, for a
   given set of calibrated parameters

   2/ Apply the Butterworth and Kernel filters . Do this a
   number of times (example 1000 times); because, the
   effects of shocks may be nonlinear, you must look at the
   Generalized impulse response functions (GIRF)

   3/ Compare the population of cross-correlations
   between the housing/residential investment and GDP
   with the cross-correlation you find when using the
   statistical data.
Alternative methodologies for
   Problem with the measures of brevity, violence,
   steepness : you do not know which variables are at play
   to account for the observed assymetries (credit
   constraints? Capacity constraints, labour markets ?).

   Alternative models including transition variables
  - Deterministic models such as TAR or STAR models
  - Stochastic models such as Markov Switching models
   with endogenous probability of transition.
Le cours s’articule autour de trois points:

Paper by Laurent Ferrara
 and Olivier Vigna
Choice of the 2-step version of the
HP filter
  Motivation : why using this filter if other filters yield
   similar turning points? Are there any robustness study
   elsewhere in the literature?

  One question : to which extend can we say that turning
   point in the housing sector are causing those observed
   in the GDP? May be we are simply detecting common

  Something original : the use of Confidence indicator in
    the building sector (perception of the activity by the
    housing industrials)  capture the supply side of the
    housing market
Long-run analysis (1)
  Most interesting part of the paper, but no rigourous
   statistical analysis to test the arguments. This seems a
   promising area of research because it relies on the
   fundamentals of then housing markets

  The conclusions challenges those of the IMF. While the
   authors argue that the movements in the housing prices
   and investment were smaller in magnitude as compared
   with the other european countries (Spain, the UK,
   Ireland), the IMF finds that France was among the
   countries with highest overvalued house price (20%)
   and a housing ratio investment significantly above the
   historical trend.
Long-run analysis (2)
  The IMF concludes that France was among the countries
   that should experience the largest decrease in the
   housing prices due to the the « greatest exuberance » in
   the house price.

  It would be interesting to see how the authors’ arguments
     can be corroborate by a statistical analysis and why their
     depart from the IMF findings.
Thanks for your attention