Transcript Evaluating the Accuracy and Rationality of UK Property Forecasts

.
Dr. Fotis Mouzakis
Dr. Papastamos Dimitrios
Prof. Simon Stevenson
Purpose
 This study examines the rationality and momentum in forecasts for rental,
capital value and total returns for the real estate investment market in the
United Kingdom compiled by the IPF
 We adopt an innovative approach of a 3-dimensional panel data
estimation, including momentum, macro-economic determinants and
structural variation (fixed-effects) over:
 Individual forecasters (cross-section)
 Separate target years (cross-section)
 Horizon of re-issuing forecasts for the same target by each forecaster (time-
series)
 The empirical model includes the effect (to forecast errors) of:
 Macro economic conditions including economic growth and investment risk
 Momentum in revising forecasts
 An approach which improves missing explanatory aspects of previous approaches
and improves under-specification bias (omitted variable estimation bias)
Literature Review
 Tsolacos (2006) &McAllister et al. (2008) studies examined the
accuracy of the IPF Consensus Forecasts for 1999-2004 period, thereby
limited to broadly strong market conditions.
 Bond & Mitchell (2011) also consider the IPF data, although in a
different context. Their analysis compares the forecasting accuracy of
the IPF Consensus Forecast for total returns versus implied forecasts
derived from total return swap contracts.
 The results, interestingly, show that for a one-year horizon, the
derivatives based implied forecasts display greater accuracy than the
consensus professional forecasts for total returns.
Data
 The data used in this study is provided by the Investment Property Forum (IPF) for the
period 1999-2011.
 The total sample consists of 69 Property forecasters
 22 property advisors, 26 fund managers and 21 equity brokers
 The forecasts provided are for growth rates/returns for the IPD ‘All Property’ Indices for
Rental, Capital Value Growth and Total Returns up to 3-year ahead period in quarterly
basis.
 The benchmark reference point is the return for the relevant annual index produced by the
Investment Property Databank (IPD) who are the primary index provider in the
commercial real-estate market in the United Kingdom.
 The macroeconomic variables that are used in this study are the gross domestic product
(GDP) and the default spread (DS). All of these variables refer to the UK economy and was
obtained from Datastream for the period 1999-2011.
Descriptive Statistics of Accuracy of the Consensus Figures
Descriptive Statistics of Accuracy of the Consensus Figures
Rationality & Momentum of Property Forecasts
 We apply the Holden & Peel (1990) approach to consider the rationality (i.e.
bias and efficiency) of property market related forecasts. The significance of
the constant, τ , indicates biased forecasts.
 Model Specification:
Results of Bias & Momentum
Formal tests of bias in the case of capital and total returns aren’t possible due
to non-stationarity issues
This is actually the result of extremely high momentum, with the coefficients
relating to lagged forecast errors being in excess of 0.9, indeed some evidence
of explosive forecast series (with b>1)
Behavioural Analysis of Property Forecasters
 We also run tests to see if forecasters are affected by general economic
and property market conditions present at the time of forecast
 We incorporate into the analysis GDP, Default Spread representing
property market conditions
Behavioural Analysis of Forecasters – Rental Growth
Target Year Fixed Effects – Rental growth
Horizon fixed effects – Rental growth
1.60%
0.30%
1.40%
0.20%
1.20%
0.10%
1.00%
0.00%
1
0.80%
2
3
4
5
6
7
8
9
10
-0.10%
0.60%
-0.20%
0.40%
-0.30%
0.20%
-0.40%
0.00%
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009



Excessively pessimistic during the strong years
1999-2004
Optimistic prior to down-turn 2005-2007
Too pessimistic in 2009 – overestimated
down-turn
-0.50%




Alternating direction of corrections when they
re-issue the forecast
Stronger corrections as forecasting horizon
shortens
Strong negative correction (more optimistic)
in 3 and 6 quarter horizons
Strong pessimistic (conservatism) final
correction in last quarter
Behavioural Analysis of Forecasters – Capital Growth
Target Year Fixed Effects – Rental growth
Horizon fixed effects – Rental growth
1.20%
2.00%
1.00%
1.50%
0.80%
0.60%
1.00%
0.40%
0.50%
0.20%
0.00%
0.00%
1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
-0.50%
-1.00%

Similar pattern of correction over the historic
years covered similar to rental growth
-0.20%
1
2
3
4
5
6
7
8
9
10
-0.40%
-0.60%




They start from negative (conservative) bias in
long horizons
They reduce negative bias (optimistic
corrections) as they re-issue forecasts
At 5q horizon they switch to negative
corrections (pessimistic corrections-with
exception 2q)
Strong final conservative correction in last
quarter – a final attempt to reduce positive
bias
Key Findings

Forecasters tend to exhibit optimistic behaviour during periods of market underperformance and vice versa.
This is consistent with the finding that they tend to exhibit greater forecast error when the property market is
underperforming and vice-versa.

The empirical findings show high levels of momentum in the forecasts, with highly persistent forecast errors.

Forecasters tend to avoid ‘big numbers’ and display conservatism

Property forecasters tend to update their previous forecast:



The results also indicate that forecasters are affected by adverse conditions


Capital value and rental growth: over-pessimistic during the strong years and most during the adjustment of
2009
Property forecasters tend to be affected by the general economic conditions at the time the forecasts are
made, including



Capital value: upwards in long horizons and downwards in short
Rental growth: Increasingly as horizon falls, but with fluctuations
Economic growth
Risk premiums (default spread)
The newly tested methodology (3d-panel) delivered statistically robust results and indications for the need to
account simultaneously for time series and cross-section variations