normally distributed residuals. Relying on the Schwarz Bayesian Information Criteria (SBC),
seasonal dummies are not included in any of the models.
The fundamental variables that are included in the models are selected by the block
exogeneity test using Sim’s small-sample corrected LR test values. The multiple variable
models for offices and industrial property incorporate nominal interest rate (IRN), interest rate
spread (S), economic sentiment (SE) and the default risk premium (D) as the fundamental
variables, whereas the models for apartments and retail property also include GDP, the
inflation rate (INF) and the real interest rate (IR) instead of the nominal one.3 The economic
sentiment, GDP and inflation cater for the expectations concerning real cash flow growth,
whereas the other fundamentals in the model (and the inflation rate, to some extent) represent
the current and expected future movements in the discount factor.
The direction of the possible Granger causality is tested by a standard F-test to
examine the existence of lead-lag relations between the assets. The multiple variable models
are also used to derive the impulse responses of real estate returns to unanticipated changes in
the fundamentals and in the real estate returns themselves. This paper employs the
‘generalized’ impulse response function developed by Pesaran and Shin (1998). The
generalized impulse responses (GIRFs) do not require orthogonalization of shocks and is
invariant to the ordering of the variables in the VAR. This is desirable, since the theory does
not give clear guidance as to which of many possible parameterizations one should use in the
traditional impulse response analysis. Furthermore, we estimate and graph 90% confidence
bands for the impulse responses by performing 100 bootstrap replications for the residuals.
Bootstrapping is conducted to diminish the problem of relatively small number of
observations on the results. Finally, the lag length in the models is decided by SBC. SBC
tends to select relatively parsimonious models. This is desirable in this analysis due to the
relatively small number of observations.