Determinants of Cross-Sectional Variation in Discount Rates, Growth Rates and Exit Cap Rates

Article excerpt

This study investigates the determinants of key input variables in valuers' discounted cash flow models used for estimating market values for offices. Data from 599 valuations in 2000 from Stockholm, Gothenburg and Malmo are used to explain variation in discount rates, expected growth rates in net operating income and exit cap rates. Our ability to explain the relatively wide variation in appraisal assumptions with plausible covariates generates confidence in the appraisal process. This has important implications because most value and returns indices of commercial real estate worldwide are appraisal based.


Most property value series are based on appraisals (NCREIF in the United States, BOMA in Australia, IPD in the United Kingdom, Sweden and elsewhere). These series, along with rent and vacancy rate data, are the basis of most property market analysis and research. Thus, how reasonably appraisers perform these valuations is of great importance.

Valuation methods vary by country. For the seven European countries for which IPD collected data in 2001, only two, Sweden and the Netherlands, obtained more than 10% of valuations by discounted cash flow (DCF) calculations. The percentage was a full 95% in Sweden. Valuers using a DCF valuation method must specify three inputs before determining value: a discount rate for cash flows, an expected growth rate in cash flow between the present and an assumed future sales date and an exit capitalization rate for this date. In this paper, we investigate the determinants of each of these three inputs. To the best of our knowledge, no prior research on this topic has been undertaken.

In fact, there has been little research on individual property or firm appraisals or capitalization rates because individual property cross sections or time series have largely not been available. A recent exception is Clayton, Geltner and Hamilton (2001) who study 202 appraisals on 33 properties over the 1986-1996 period supplied by two Canadian real estate managers. Given the multiple appraisals of only 33 properties, these authors are concerned with the time-series appraisal smoothing issue, rather than cross-sectional determinants of valuations and valuer inputs. Given the purely cross-sectional nature of our data, we do not address the smoothing issue.

The only other studies of individual property databases are by Hendershott and Turner (1999) and Janssen, Soderberg and Zhou (2001) who both analyze transaction-based cap rates in Stockholm. Hendershott and Turner compute constant-quality cap rates over the five semiannual periods 1990.2-1992.2 using data for 403 properties. Property-specific information includes density of land use, age, property type and a measure of below-market financing. Time dummy variables are included in their model to capture the impact of changing macroeconomic variables, which allows computation of the semiannual cap rate series. Janssen, Soderberg and Zhou explain cap rates on 302 properties (largely residential) over the 1992-1994 period. The determinants are property type, age and dummy variables for four areas of the city.

In the finance literature, the most relevant paper is the cross-sectional analysis of 1,005 companies by Cho (1994). He finds that individual company price/earnings (P/E) ratios, the inverse of the cap rate, are positively related to earnings growth rates and dividend payout rates and negatively associated with the standard deviation in growth rates.

Using 2000 data on 599 office properties in the three largest cities of Sweden, we attempt to explain the variation in four variables: discount rate, expected cash flow growth, exit cap rate and long-run vacancy rate. We first lay out the valuer's framework and then describe our data set. Next we discuss hypotheses regarding the cross-sectional variation in these variables and discuss estimation issues. Finally, we report tests of the hypotheses and discuss the quantitative impacts of the determining variables. …