Valuation accuracy across Europe: a mass appraisal approach

2020, Journal of Property Research

visibility

…

description

24 pages

link

1 file

Abstract

The aim of this paper was to compare valuation accuracy of eight European markets, using the same time period, data source and methodology. The emphasis was placed on the accuracy of held properties because previous studies showed that sold properties tend to be valued closer to the market. Real sales data was used to derive hedonic sale prices. The Heckman correction was employed to correct for sample selection bias. A comparison of simple differences between actual valuations and fitted prices showed that valuations were on average below fitted prices in all countries except the Netherlands, indicating a possible overvaluation problem of held properties in Europe. A comparison of the absolute difference showed that the Netherlands and Switzerland displayed the highest valuation accuracy. Italy and Sweden on the other hand were the markets with the lowest median valuation accuracy and largest spreads of observations. All countries, except Sweden, had a majority of observations within an absolute difference of 20%. The two most interesting conclusions from the analysis were that Germany and Switzerland did not differ significantly from other markets in terms of valuation accuracy and that Sweden was consistently the market with the lowest valuation accuracy.

Figures (14)

PISUre 1PISule 42 Alsplays Ule alludl lOldal TELUTTIS OF Cleilt ValudUOll Das€G IVWLOWL PLOPerry indices. The indices measure the performance of unlevered, directly held real estate. Total return patterns differed drastically between countries. The highest average annual total returns over the years 2002-2013 were found in France (9.1%) and the UK (7.6%). The standard deviations were 6.9 and 11.6 percentage points, respectively. Switzerland and Germany on the other hand had lower annual average returns (6.0% and 3.2%, respectively) but also experienced much lower volatility. The standard deviations were only 0.8 and 1.5 percentage points, making them the markets with the highest return per unit of standard deviation. Rather than market movements, the observed differences could be the result of cross-country differences in valuations. Valuation accuracy, the difference between appraised values and market prices, has long been the subject of debate among academics, practitioners and the Courts (Lizieri & Venmore- Rowland, 1991; McAllister, 1995). If valuations failed to accurately mirror underlying

market in contrast to other countries (ocnnaldt & sebastian, 2U12; Crosby, 2UU/). ni variation among different valuers (Boyd & Irons, 2002; Brown, 1992; Crosby et al., [he term valuation accuracy has occasionally been used to refer to issues of valuation 1998: Lizieri & Venmore-Rowland, 1993). However, valuation accuracy in this paper is definec as the difference between appraised values and market prices (Boyd & Irons, 2002). The easiest way to assess valuation accuracy would be to compare the price achieved market to a valuation at the same moment in time. While there are some sma studies with such a set-up (see Parker, 1999), this approach is problematic in real | in the 1 case ife. In most cases, a valuation is carried out in lieu of a transaction and the sale that the valuation emulates remains hypothetical (Gallimore & Wolverton, 2000). Even i f both numbers were available simultaneously, it is unlikely that the valuation has been derived independently and is unaffected by information from the transaction proceedings. 7 known as the influence window (Walvekar & Kakka, 2020). This is

Table 1. Number of observations per country. aii MAE UVitivi PSEAGLNS LT 9: LiIth AVALIAUIN UACAIAOLE LLIVILU Lil 2YU1iY ALINE 11 AUUILIVildi Udlda could be obtained. At the end of 2013, the final year of the data set, the regional MSCI databanks contained 168,394 entries of domestic properties (Table 1)Table 6. By far the largest share, 63.3%, came from the UK which was due to historical reasons, as the UK was the first country in which data collection started in the 1980s. Since then, services have been expanded to numerous countries around the globe. The next largest share of observations (10.6%) came from France and the other six countries made up the remaining 26.1% all together. The analysis was limited to office, retail and industrial properties. Residential properties and properties in the category ‘other’, which includes anything from hotels and nursing homes to parking spaces and cinemas, were removed in order to reduce heterogeneity. This eliminated 38.9% of entries. The remaining 102,919 entries were split into 817,063 annual observations of sales and valuations. The earliest year of available data varied between 1980 for the UK and 2002 for Italy. In ordet to ensure comparability, observations prior to 2002 were dropped. This eliminated 56.8% of observations, mostly due to the omission of observations from the UK dataset. Finally. assets under development and assets with partial sales or purchases were dropped. At this point, the preliminary dataset consisted of 319,736 observations across all eight countries. Some observations only included the minimally required variables for the perfor- mance analyses. Non-essential information, such as age or market rents, were not always recorded. Excluding observations which do not contain a full set of variables required for the subsequent regressions, reduced the final dataset to 225,033 observations (Table 1). Just as in the uncleaned dataset, the majority of observations came from the UK. However, the UK’s share of observations dropped from 63.3% to 44.9% during the cleaning process. 21.5% of observations in the cleaned dataset came from France and 11.9% from the Netherlands. The other five countries made up the remaining 21.7%. The country with the smallest number of observations was Spain (1.2% = 2,751 observations).

The first step of the analysis consisted of deriving a standard regression model that could be applied to valuations and sale prices in all countries. Table 2 displays the results of the valuation regressions. The explained variable was the natural log of value per square metre, In(Value). All subsequent models followed a similar set up. Explanatory variables were based on theory and availability of data from MSCI. Coefficients were obtained via Ordinary Least Squares (OLS) with heteroscedasticity robust standard errors. The explanatory variables included year dummies, with the base year 2002, followed by over weasel: quanectin ll aenlesss [FRAC aed aw cerormwewres Jecdianébne TOWN has acces

Confidence Level: *10%, **5%, ***1%

Table 3. Location categories per country.

Figure 4. Simple difference between actual valuations and fitted valuations.

Table 4. Naive pricing regressions. Confidence Level: *10%, **5%, ***1%

Table 5. Heckman pricing regression models.

Confidence Level: *10%, **5%, ***1%

Table 6. Proportion of observation within an absolute difference of 30%.

Figure 5. Simple difference between actual sale prices and fitted prices.

Figure 6. Simple and absolute difference between actual valuations and Heckman Prices.

possibly Sweden, which had a comparatively large median difference in 2007. The difference between actual valuations and fitted prices in the UK and Italy seems to have increased in the years since the financial crisis. Further, it looks like property cycles were not particularly noticeable in the annual levels of valuation accuracy in Figure 7.