1
Forward-looking Enterprise Value Multiples:
An Empirical Research
KRIJN JELLE DIJKSTRA Master Thesis MScBA Finance1,
June 2013
ABSTRACT
We examine whether the accuracy of the company value estimated by the multiples valuation method, increases as forecasts are used instead of historical earnings for EV/EBITDA and EV/EBIT multiples. Secondly, we examine if the accuracy of the value estimate increases further as later-year instead of earlier-year forecasts are employed. This is also re-examined for the P/E multiple. The analysis is conducted on a sample of 1,388 companies from Germany, France, and Great Britain. Results of the Mann-Whitney U test show that the accuracy of the value estimate significantly increases as any of the three forecast years is used instead of historical earnings, for all three multiples. This accuracy does not increase further as later-year instead of earlier-year forecasts are employed, for all three multiples.
Keywords: corporate valuation, forward-looking multiples JEL classifications: G12, G30
1 University of Groningen, Faculty of Economics and Business, MSc Business Administration, Finance. Student number: S1588397
2
I. Introduction
The multiples valuation method is one of several methods that can be applied in order to estimate the value of a company. It is often used to supplement other valuation methods, such as the discounted cash-flow method. Schreiner and Spremann (2007) define a multiple of a company as the ratio of a market price variable (like market capitalization or enterprise value) to a particular value driver of that variable (e.g. earnings or revenues). The multiples valuation method roughly consists of four steps. Firstly, one needs to select the kind of value (like enterprise or equity) that is of interest, and select a corresponding value driver. Secondly, a group of companies that is comparable to the target firm is gathered, from whose multiples a single, aggregate multiple needs to be estimated. Step three consists of estimating the aggregate multiple, according to a preferred measure of central tendency. This aggregate multiple is assumed to be representative for all those comparable companies, and therefore also for the target company. The final step is to estimate the value of the target company, by multiplying the aggregate multiple with the value driver of the target company. So, what the multiples valuation method basically does is base the value of the target company on how similar companies are currently valued by the market (Damodaran, 2002).
3 Multiples using historical earnings are called trailing multiples, and multiples using earnings forecasts are called forward-looking multiples. Research on the accuracy of the value estimate resulting from the use of forward-looking multiples has shown that the accuracy of the value estimate indeed increases as forecasts are used instead of historical (or trailing) earnings (Kim and Ritter, 1999; Lie and Lie, 2002; Liu, Nissim, and Thomas, 2002; Schreiner and Spremann, 2007). The accuracy of the value estimate was also found to increase further as later-year forecasts are used rather than earlier-year forecasts (Kim and Ritter, 1999; Liu et al., 2002 Schreiner and Spremann, 2007). However, these findings are all based on the price to earnings (P/E) multiple. Whether the accuracy of the target company value estimate made by the multiples valuation method using EV/EBITDA or EV/EBIT multiples increases as forecasts of earnings are used in lieu of historical earnings has never been examined. This therefore is the main research question of this paper. Next to that, it is also examined whether the accuracy of this value estimate further increases as later-year forecasts are used instead of earlier-year forecasts. These two questions are also re-examined for the P/E multiple.
4 add to the literature on the multiples valuation method in general, and, more specifically, will fuel the discussion on the application of forward-looking enterprise value multiples. These are much used in practice, but remain underexposed in the literature.
This empirical research is applied on a sample of 1,388 companies from Europe’s three largest countries; Germany, France and Great-Britain. Enterprise value and market capitalization are of ultimo 2011, where enterprise value is defined as the market value of equity plus the book value of debt, minus cash and cash equivalents. The EBITDA, EBIT, and net earnings used to calculate the trailing multiple are also from 2011. The EBITDA, EBIT, and net earnings forecasts from 2012, 2013, and 2014 make up the one-year, two-year, and three-year ahead forward-looking multiples. First, we will examine the total sample, to test if the accuracy of the valuation errors increases as forward-looking multiples are used. This implies an aggregate multiple is estimated from the total sample (Liu et al., 2002). Second, the analysis is performed on a country level, where the multiple is estimated from each country. After that, the sample is subdivided into industries according to the NACE Rev. 2 codes. An aggregate multiple is now estimated per industry. This is the most detailed part of the analysis. Value estimates should now be more accurate than for the earlier samples. After the valuation errors of the separate industries are calculated, they are examined by pooling them into a single distribution. We evaluate companies from financial industries separate from the rest of the companies, and thus end up with two separate industry subsamples (Lie and Lie, 2002). The aggregate multiple is always estimated out-of-sample, implying that the target company is not part of the group from which the aggregate multiple is estimated.
The main research question of this paper is backed by our results. For the EV/EBITDA multiple, the median of the absolute valuation errors shows a decrease for the total sample and every subsample, as forecasts are used instead of trailing data. This implies an increase in the accuracy of the estimate of company value. The results of the one-sided Mann-Whitney U test show that the use of any of the forward-looking multiples in lieu of the 2011 trailing multiple produces significantly more smaller valuation errors for all examined samples, besides for the non-financial industry subsample. The median of the absolute valuation errors of the EV/EBIT multiple also show a decrease as forecasts are used instead of trailing data, for the total sample, the country subsamples, and both industry subsamples. The results of the one-sided Mann-Whitney U test, are significant for the total sample and all the country and industry subsamples. This shows that the use of forecasts from 2012, 2013, or 2014 instead of trailing data always leads to significantly more smaller valuation errors.
5 one-sided Mann-Whitney U test never reveals a difference in the accuracy of the value estimates as later-year forecasts are used instead of earlier-later-year forecasts, for the total sample or the country subsample. Employing the industry subsample does not improve the relative performance of these forward-looking multiples.
The results of the analysis on the P/E multiple are in line with the earlier findings (Kim and Ritter, 1999; Lie and Lie, 2002; Liu et al., 2002; Schreiner and Spremann, 2007) that the use of earnings forecasts instead of trailing earnings increases the accuracy of the value estimate of the P/E multiple. For the total sample, the country subsamples, and the industry subsamples, the one-sided Mann-Whitney U tests finds a significantly larger amount of smaller valuation errors as earnings from any of the three forecast years are used in lieu of trailing earnings. However, our results do not unanimously support the finding that the use of later-year forecasts produce more accurate estimates of value than the use of earlier-year forecasts for the P/E multiple (Kim and Ritter, 1999; Liu et al., 2002; Schreiner and Spremann, 2007). The median of the absolute valuation errors does show a decrease as later-year forecasts are used, implying an increase in the accuracy of the value estimate. However, the results of the one-sided Mann-Whitney U test turn out to be significant only once for the total sample and once for the French subsample, showing that when distributions with the absolute valuation errors are compared, the later-year ones contain a significantly higher amount of smaller valuation errors only twice. Again, the relative performance of the forward-looking multiples does not change as the industry subsamples are employed.
The remainder of this paper is organized as follows. The next section discusses the literature on multiples. In section three the methodology is discussed, and section four contains the data description. In section five the results of the research questions are presented and discussed. The conclusion is presented in the final section. Due to the extensiveness of the descriptive statistics and the results, some of these are presented in the appendix.
II. Literature
This section discusses the existing literature on forward-looking multiples. Table I, on page 7, presents an overview of the research approach of these papers.
6 two-year ahead forward-looking multiple, with a median absolute valuation error of 22.2%. The one-year ahead forward-looking multiple and trailing multiple follow, with a median absolute valuation error of respectively 40.1% and 52.1%. Moving from the trailing to the two-year ahead forward-looking multiple, the amount of absolute valuation errors smaller than 15% increases from 18.9% to 27.3% to 28.0%. Liu et al. (2002) compare the forward-looking earnings to price (inverted P/E) multiple to, among others, the per share book value to price multiple, the EBITDA/P multiple, and the sales to price multiple, employing a sample of 19,879 observations between 1982 and 1999. They perform their analyses on the total sample, and on industry level. Mostly reviewing the relative performance of the multiples, they establish that forward-looking E/P multiples outperform all other trailing multiples, and that the performance improves as the earnings forecast horizon is lengthened from one-year to two-year to three-year out. So, even though the methodology of these two papers largely differs, they both advocate the use of earnings forecasts, and base this on equity based multiples.
The valuation bias and accuracy of the value estimate of the trailing and one-year forward-looking P/E multiple are examined by Lie and Lie (2002). To assess the bias of the multiples, they use the median in favour of the mean, so as to mitigate for outliers. Valuation errors for the total sample show that the use of earnings forecasts improve the performance of the P/E multiple. The median of the absolute valuation error decreases from 33.4% to 28.3%, and the percentage of absolute valuation errors below 15% rises from 25.5% to 31.0%. The bias is very small, it decreases from 0.002 to -0.001, which implies the methods yield a negligible valuation bias. Next, the valuation errors are analysed as they are pooled into either of two categories, financial and non-financial, and then again subdivided according to three levels of both size and earnings. Lie and Lie make a distinction between financial and non-financial companies because financial companies are easier to value, due to the liquid nature of their assets. The subdivisions cause the bias of the trailing P/E multiple to increase, but for nine out of twelve this bias decreases again as the earnings forecasts are employed. Now, valuation accuracy is only measured by the percentage of absolute valuation errors below 15%. This percentage increases in all twelve subsamples as forecasts of earnings are used. The value estimates are more accurate for financial companies, but the trends are similar in both groups. Overall, Lie and Lie consistently find that the one-year forward-looking multiple produces more accurate value estimates than the trailing multiple does.
7 Table I. Overview of relevant literature
Sample description Other relevant properties Relevant multiples examined Methods of valuation accuracy analysis Rank of performance of forecast years Kim and Ritter (1999)
143 company IPOs from the US from 1992 and 1993. Firms must have positive earnings. Earnings forecast data from 1993 or 1994. All data comes from the Securities Data Company. P/E ratios above 100 are maximized at 100. Analysis on industry level
Unknown which earnings forecast measure is used. Aggregate multiple is estimated using the geometric mean. Valuation error is calculated as the log of (predicted/actual).
Trailing P/E, one-year and two-year forward-looking P/E.
Mean absolute valuation error. Median absolute valuation error. Percentage of absolute valuation errors smaller than 15%.
1. 2-year out forecast 2. 1-year out forecast 3. Trailing multiple
Lie and Lie (2002)
8,621 US firms from compustat. Firms must have positive earnings. All financial data is from 1998. Earnings forecast data is from 1999, from the IBES database. No info on outlier handling. Analysis on entire sample, and industry level. Separate analysis of financial and non-financial company.
Median earnings forecast is used. Aggregate multiple is estimated using the median. Valuation error is calculated as the log of (predicted/actual).
Trailing P/E and one-year forward-looking P/E.
Mean absolute valuation error. Median absolute valuation error. Percentage of absolute valuation errors smaller than 15%. Interquartile range of absolute valuation errors.
1. 1-year out forecast 2. Trailing multiple
Liu et al. (2002)
19,879 US firm years, from 1987 to 2001. Firms must have positive earnings. One-year and two- year out earnings forecast from IBES. The three- year out earnings are estimated using a growth rate. Top & bottom 1% of observations of each variable are deleted. Analysis on entire sample, and industry level.
Mean earnings forecast is used. Aggregate multiple is estimated using the harmonic mean, and the median. Valuation error is calculated as the actual price less the predicted price, scaled by the actual price.
Trailing inverted P/E, one-year, two-one-year, and three-year forward-looking inverted P/E.
Mean valuation error. Median valuation error. Interquartile range, 90% - 10% range, and 95% - 5% range of valuation error.
1. 3-year out forecast 2. 2-year out forecast 3. 1-year out forecast 4. Trailing multiple
Schreiner and Spremann (2007)
592 firms from 1996 to 2005, from the European Dow Jones STOXX 600 index. Firms must have positive earnings. One-year and two-year out earnings forecast from the IBES database. No info on outlier handling.
Mean earnings forecast is used. Aggregate multiple is estimated using the median. Valuation error is calculated as the actual price less the predicted price, scaled by the actual price.
Equity based trailing, one-year and two-one-year forward-looking multiples, with Sales, EBITDA, EBIT, EBT, and net earnings
Median absolute valuation error. Percentage of absolute valuation errors smaller than 15% and 25%. Interquartile range of absolute valuation errors.
1. 2-year out forecast 2. 1-year out forecast 3. Trailing multiple
8 With the median of the absolute valuation error and the percentage of absolute valuation errors below 15% as indicators of valuation accuracy, Schreiner and Spremann find that the two-year forecasts perform the best for every multiple examined. The one-year forward-looking multiple always comes second, and the trailing multiple ends up last of the three. The median absolute valuation error of the P/E multiple decreases from 29.3% to 24.4% to 21.6%, and the percentage of absolute valuation errors below 15% increases from 30.9% to 36.1% to 39.5%, as they move from historical earnings to an increase in the forecast horizon of one and two years. Comparing the multiples themselves, the P/EBT performs the best with the most accurate valuations, but is very close to the P/E multiple. Next are the P/EBIT and P/EBITDA, and last is the price to sales multiple.
A summary of the sample requirements, the methodology of analysis employed, and the results of the aforementioned literature is presented on page 7, in table I. It becomes apparent that the methodology of every paper is different from the next. Samples differ in size and origin, aggregate multiples are estimated differently, and valuation errors are calculated in various ways between the papers. There is, however, consensus on how to analyse the valuation errors. Most used are the median of the absolute valuation error, the interquartile range of the absolute valuation errors, and the percentage of absolute valuation errors below 15%. There is also conformity in the conclusions of the research: Using earnings forecasts for P/E multiples increases the accuracy of the value estimate, and using later forecast years over earlier forecast years further increases the accuracy of this estimate.
9
III. Methodology
This paper examines whether the accuracy of the company value estimated by the multiples valuation method increases as forecasts of earnings are used instead of trailing earnings. This is examined for EV/EBITDA, EV/EBIT and P/E multiples. Secondly, whether the accuracy of these value estimates increases as later-year rather than earlier-year forecasts are used, is also examined.
The multiple of a company is calculated by either dividing its 2011 enterprise value by its EBITDA or EBIT, or by dividing its 2011 market capitalization by its net earnings. When the 2011 trailing value driver is used, the multiple is called the trailing multiple, when the forecasts of 2012, 2013, or 2014 are used, the multiple is deemed a forward-looking multiple.
Initially, the analysis is performed on the total sample, and on the country subsample. This means that the aggregate multiple is estimated from the multiples of all the companies in the total sample (Liu et al., 2002), and after that from all the companies per country subsample. This is not useful in practise, but it allows us to examine the aggregate patterns of the different multiples. Finally, the analysis is performed on an industry level. Companies are grouped according to their NACE Rev. 2 code, and the aggregate multiple is estimated from the multiples of the all companies in the same industry. This final part of the analysis is much more similar to the practical application of the multiples valuation method. The aggregate multiple can now considered as a benchmark, because it is distilled from companies that are similar to the company that is being valued. Alford (1992) established that industry membership is an effective criterium for selecting comparable firms. For this subsample, we also require the companies to have data for all three value drivers in 2011, 2012, and 2013. This restriction means that the distributions of the first three years of every multiple contain exactly the same companies. So the value estimate should become more accurate, and the increased accuracy as forward-looking multiples are used is examined for the same companies. We therefore expect the results of the industry subsamples to give a better perspective of the added accuracy of using forward-looking multiples.
10 the target company. Ideally, an aggregate multiple is estimated from five to ten companies that are similar to the target company (Cooper and Cordeiro, 2008), but due to its complexity we are unable to estimate aggregate multiples from a fixed number of companies.
The accuracy of a value estimate generated with the multiples valuation method is expressed in terms of a valuation error. This error is calculated by subtracting the aggregate multiple (which represents the estimated value) from the actual multiple (which represents the actual value) of the target company, and scaling this by the actual multiple of the target company (Liu et al., 2002; Schreiner and Spremann, 2007). Thus, the valuation error represents the percentage difference between the actual value of the target company and the value estimated by the multiples valuation method. This implies that the smaller that percentage is, the more accurate the estimate of company value is. Per multiple, per year, these valuation errors are pooled into a distribution. The median of this distribution is an indicator of the valuation bias of the multiple, which shows whether the multiple employed undervalues or overvalues a company (Lie and Lie, 2002). A median valuation error of zero means that there is no bias. Next, the absolute values of the valuation errors are computed, and again pooled into distributions per multiple, per year. The difference in the accuracy of the value estimates resulting from the use of trailing and forward-looking multiples can then be examined by comparing the distributions with the absolute valuation errors of the multiples. To examine the research question, the differences between the distributions with the absolute valuation errors resulting from the multiples valuation method are examined. The differences between the distributions are analysed by comparing their median, the interquartile range, and the percentage of valuation errors that is smaller than 15%. These three methods are also used in the previous literature (Kim and Ritter, 1999; Lie and Lie, 2002; Schreiner and Spremann, 2007). As we expect the accuracy of the value estimates to increase as (later-year) forecasts are used, we expect the valuation errors to decrease. So, as (later-year) forecasts are used for the multiples valuation method, we expect the median to decline, the interquartile range to become smaller, and the percentage of valuation errors that is smaller than 15% to increase. A move of these measures in the aforementioned directions are indications that the accuracy increases.
11 • H0; The two population locations of the absolute valuation errors are equal.
• H1; The population of the later-year multiple contains a significantly higher amount of smaller
valuation errors than the earlier-year multiple.
First, the trailing multiple is compared with each of the forward-looking multiples (2011 vs. 2012, 2011 vs. 2013, 2011 vs. 2014). After that, the earlier-year forward-looking multiples are compared with the later-year forward-looking multiples (2012 vs. 2013, 2012 vs. 2014, 2013 vs. 2014). The one-sided Mann-Whitney U test is applied with a 95% confidence level. This means that a result is significant if the Z-score is -1.645 or lower, leading to a rejection of that particular null hypothesis. A significant result thus means that the distribution of the later-year multiple contains a significantly higher amount of smaller valuation errors, which in turn leads us to conclude that using this multiple leads to more accurate estimates of target company value than the use of the earlier-year multiple.
IV. Data
The entire dataset is gathered from the Orbis database, which is provided by ‘Bureau van Dijk’. It is comprised of all companies with a known enterprise value of ultimo 2011, based in any of the three economically largest countries in Europe: Germany, France, and Great Britain. Enterprise value is defined as the market value of equity plus the book value of debt, minus cash and cash equivalents. The EBITDA, EBIT, and net earnings data used for the trailing multiple is also from 2011. For the forward-looking multiples the median of the earnings forecasts are used rather than the average, because the median mitigates the effect of outliers (Lie and Lie, 2002). Forecasts of EBITDA, EBIT, and net earnings are gathered from 2012, 2013, and 2014. Below on page 12, table II shows how the available data has been tooled into workable samples.
12 value driver. As financial companies are easier to value due to the liquid nature of their assets (Lie and Lie, 2002), they are analysed separately. They are defined as companies whose NACE code starts with 6. A disadvantage is that the financial companies subsample is relatively small, with only 88 members.
Table II. Sample preparation
Initial sample: All companies with a known EV in 2011 2948 - Companies with negative earnings in 2011 -762 - Companies with negative earnings forecasts -685 - Companies in top and bottom 1% of every value driver -113
Total number of companies in the initial sample 1388 ( Of which German 401 )
( Of which French 371 ) ( Of which British 616 )
- Companies with both no 2012 and 2013 forecast data -665 - Companies with too few industry peers or no membership -23
Total number of companies in industry sample 700 ( Of which non-financial 612 )
( Of which financial 88 )
This table depicts the preparation and final composition of the sample and subsamples.
The summary statistics of the sample are shown below, in table III on page 13. The descriptives of the companies’ enterprise values and market capitalizations, presented in panel A of table III, show that the values of the larger companies have a strong influence on the composition of the dataset. This becomes apparent since the means are much larger than the 75th percentile. This is also observed for the EBITDA, EBIT, and net earnings of every year (see appendix, table AI). The initial sample is thus heavily skewed to the right. The degree to which the data is skewed is smaller when we look at the company multiples (panel B of table III), but still the mean is larger than the median, which indicates a positively skewed distribution.
13 distributed. Table AII in the appendix depicts the Jarque-Bera statistics of the industry subsamples. These test results are significant for all years and multiples, which implies that the absolute valuation errors are also non-normally distributed for these subsamples. This means that only non-parametric formal tests, such as the Mann-Whitney U test, can be performed on the distributions with the absolute valuation errors.
Table III. Summary statistics
A. Descriptive statistics of company values. Numbers should be multiplied by € 1,000,000
Mean Median Q25 Q75 N
Enterprise value ultimo 2011 1,599 146 35 835 1388
Market capitalization
ultimo 2011 1,227 115 35 667 1388
B. Descriptive statistics of the company multiples.
Mean Median Q25 Q75 N EV/EBITDA 2011 9.26 6.3 4.16 10.14 1384 2012 7.13 5.87 4.14 8.31 870 2013 6.48 5.48 3.77 7.47 952 2014 5.88 5.00 3.46 6.88 871 EV/EBIT 2011 15.51 9.42 6.07 14.49 1388 2012 9.70 8.25 5.75 11.35 893 2013 8.35 7.32 5.10 9.91 975 2014 7.42 6.63 4.53 8.97 890 P/E 2011 28.10 12.06 8.07 18.43 1388 2012 14.59 11.64 8.25 15.91 861 2013 11.82 10.19 7.37 13.21 936 2014 9.97 8.49 6.36 11.59 856
C. Jarque-Bera test for non-normality, on the distributions containing the absolute valuation
errors. Significant results prove non-normality, and are marked with an (*).
EV/EBITDA EV/ EBIT P/E
2011 182,294* 122,309* 5,063,334*
2012 694,988* 612,833* 1,452,753*
2013 997,899* 917,173* 2,604,577*
14
V. Results
First, the results of the analysis on the total sample and on the separate countries are discussed. Next, the outcomes of the industry level analysis are discussed. The interquartile range and the percentage of absolute valuation errors below 15% of both the total sample and of the country and industry subsamples are presented in the appendix, due to the extent of the results.
Table IV on page 15 shows the results of the analysis on the total sample and on the separate countries. Panel A depicts the bias of the used multiples. The medians of the different valuation errors suggest that none of the employed multiples yield biased estimates for these samples. Panel B depicts the median of the absolute valuation errors, with the results of the one-sided Mann-Whitney U tests below them in brackets. Focusing on the median of the absolute valuation errors of the total sample first, there is a large improvement in the accuracy of the value estimate when moving from the 2011 trailing to the one-year forward-looking multiple. For the EV/EBITDA multiple, the median of the absolute valuation errors decreases by 9%. For the EV/EBIT and P/E multiple, it decreases by 8.1% and 7.2%, respectively. The accuracy increases further as forecasts from 2013 and 2014 are used instead of the trailing data from 2011. However, the improvement in the accuracy of the value estimate of moving from the 2012 forward-looking multiple to the 2013 and 2014 forward-forward-looking multiple is much smaller. The improvement of moving from the use of 2012 forecast data to 2014 forecasts, as measured by a decrease in the median of the absolute valuation error, is only 0.4% for the EV/EBITDA multiple. For the EV/EBIT and P/E multiple this improvement is larger, as the medians decrease by 2.7% and 5.1%, respectively.
15 Table IV. Valuation errors for the total sample
A. Median of the valuation errors. Here, the median is an indicator that shows to which extent the valuation methods are biased,
where 0 indicates no bias. DE stands for Germany, FR for France, and GB for Great Britain. Total depicts the entire sample.
EV/EBITDA EV/EBIT P/E
DE FR GB Total DE FR GB Total DE FR GB Total
2011 -0.001 -0.006 0.001 0.000 -0.001 0.000 0.000 0.000 0.000 0.000 0.000 0.000 2012 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 -0.001 0.001 2013 0.006 0.000 0.001 0.000 0.000 0.000 -0.001 0.000 -0.001 0.002 0.000 0.000 2014 0.003 0.001 0.005 -0.001 0.000 0.002 0.002 0.000 0.004 0.000 0.000 0.000
B. Median of the absolute valuation errors, and Z-scores from one-sided Mann-Whitney U test (in brackets). These results show if the three
looking multiples produce more accurate estimates of target company value than the 2011 trailing multiple, and if later-year forward-looking multiples produce more accurate estimates of target company value than earlier-year forward-forward-looking multiples. A decrease in the median as we move to (more) forward-looking multiples implies an increase in the valuation accuracy. The difference between the distributions of the absolute valuation errors of the former multiple and the latter multiple is significant at a 95% confidence level if the Z-score is (-1.645) or lower, implying that the latter provides a more accurate value estimate. Significant results are marked with an (*).
EV/EBITDA EV/EBIT P/E
DE FR GB Total DE FR GB Total DE FR GB Total
16 On a country level all three multiples show similar patterns for all three methods of analysis. There is a substantial increase in the accuracy of the value estimate as we move from the trailing to forward-looking multiples, but a less distinct improvement as we move from earlier to later-year forecasts. Again, the P/E multiple benefits the most from using later-year forecasts instead of earlier-year forecasts.
These initial results are mostly confirmative with regard to our research questions and the literature on forward-looking P/E multiples. The use of forecast data increases the accuracy of the value estimate of the P/E multiple and, as hypothesized, this is also the case for the use of forecast data for EV/EBITDA and EV/EBIT multiples. We also find a further increase in the accuracy of the value estimate as later-year forecasts are used instead of earlier-year forecasts, however this increase is smaller than with the use of forecast over trailing data.
To examine the significance of these outcomes, the results of the one-sided Mann-Whitney U test are discussed. Panel B of table IV on the previous page shows the results of the one-sided Mann-Whitney U tests between brackets. A Z-score that is lower than -1.645 implies a significant result with a confidence level of 95%, and is marked with an asterisk (*). First, the distribution with the absolute valuation errors of the 2011 trailing multiple is compared with that of each of its three forward-looking peers. For the total sample and the three countries, for all three multiples, we found significant results. Thus we reject the null hypotheses. This leads us to conclude that, for the total sample and the country subsamples, the forward-looking multiples produce more accurate estimates of company value than the trailing multiple does. This leads to an affirmative answer on the main research question of this paper, which asks whether the use of forecast data for EV/EBITDA and EV/EBIT multiples leads to more accurate value estimates than the use of trailing data does. The results of the Mann-Whitney U test also support the notion that using forecast data for the P/E multiple instead of historical earnings leads to a more accurate estimate of company value, as was also found by previous research (Kim and Ritter, 1999; Lie and Lie, 2002; Liu et al, 2002; Schreiner and Spremann, 2007).
17 EV/EBIT multiple. Concerning the P/E multiple, our results do not show strong evidence in support of findings in the previous literature (Kim and Ritter, 1999; Liu et al., 2002; Schreiner and Spremann, 2007).
Table V. Valuation errors for the industry subsamples
Non-Financial industries Financial industries
A. Median of valuation errors. Here, the median is an indicator that shows to which extent the valuation methods
are biased, where 0 indicates no bias.
EV/EBITDA EV/EBIT P/E EV/EBITDA EV/EBIT P/E
2011 0.004 0.006 0.000 -0.022 0.023 -0.021
2012 0.001 -0.006 -0.006 0.001 0.006 0.005
2013 0.000 0.002 0.001 0.000 -0.009 -0.005
2014 0.007 0.000 -0.006 -0.001 0.018 -0.002
B. Median of absolute valuation errors, and Z-scores from Mann-Whitney U test (in brackets). These results show
if the three forward-looking multiples produce more accurate estimates of target company value than the 2011 trailing multiple, and if later-year forward-looking multiples produce more accurate estimates of target company value than earlier-year forward-looking multiples. A decrease in the median as we move to (more) forward-looking multiples implies an increase in the valuation accuracy. The difference between the distributions of the absolute valuation errors of the former multiple and the latter multiple is significant at a 95% confidence level if the Z-score is (-1.645) or lower, implying that the latter provides a more accurate value estimate. Significant results are marked with an (*).
EV/EBITDA EV/EBIT P/E EV/EBITDA EV/EBIT P/E
2011 34.1% 32.7% 34.1% 45.3% 42.5% 42.2% 2012 30.6% 28.3% 32.2% 31.5% 34.0% 34.1% (2011 vs. 2012) (-0.98) (-2.45)* (-0.66) (-2.17)* (-2.21)* (-2.40)* 2013 30.3% 27.6% 26.4% 27.8% 28.0% 25.1% (2011 vs. 2013) (-1.09) (-2.53)* (-3.26)* (-3.10)* (-2.91)* (-3.38)* (2012 vs. 2013) (-0.05) (-0.10) (-2.66)* (-1.29) (-0.72) (-0.79) 2014 29.3% 26.9% 27.5% 27.0% 28.9% 27.3% (2011 vs. 2014) (-0.98) (-2.88)* (-2.80)* (-2.97)* (-2.96)* (-3.54)* (2012 vs. 2014) (-0.01) (-0.57) (-2.22)* (-1.14) (-0.37) (-1.21) (2013 vs. 2014) (-0.05) (-0.47) (-0.42) (-0.12) (-0.30) (-0.23)
18 the value estimates should improve. Furthermore, only companies that have data for all three value drivers in 2011, 2012 and 2013 are in the subsample, so the distributions with the absolute valuation errors of 2011, 2012 and 2013 consist of the same companies. We therefore expect these results to give an improved perspective of the increase in accuracy of the value estimate resulting from the use of forward-looking multiples in lieu of trailing multiples.
First, the results of the non-financial industry sample are discussed. These are on the left side of table V on the previous page. Panel A shows that the valuation bias is very small in every year, for all three multiples. Thus, the use of these multiples again does not yield biased estimates. Panel B depicts the results of the analysis of the absolute valuation errors. As 2012 forecasts are used instead of 2011 trailing data, the median of the absolute valuation errors of the EV/EBITDA multiple decreases with 3.5%, that of the EV/EBIT multiple with 4.4%, and that of the P/E multiple with 1.9%. The increase in the accuracy of the value estimate continues as later-year forecasts are used. For the EV/EBITDA and EV/EBIT multiple, the use of the 2013 forecasts causes the median to decrease with 0.3% and 0.7%, and the use of 2014 forecast data decreases it with a further 1.0% and 0.7%. Where the additional accuracy of the use of 2012 forecast data is small for the P/E multiple, the use of 2013 forecast data causes the accuracy of the value estimate to increase with 5.8%. However, using the 2014 forecast instead of the 2013 forecast causes a decrease in the accuracy of the value estimate with 1.1%.
Summarized, the median of the absolute valuation errors again implies that the EV/EBITDA and EV/EBIT multiple both benefit from the use of forecast data in lieu of trailing data, but now the increase in accuracy of the value estimate is smaller than for the total sample. Besides that, the interquartile ranges of these two multiples (appendix, table AIV) actually increase as we move from the use of 2013 forecasts to 2014 forecasts, while we would expect them to decline. The percentage of valuation errors that is smaller than 15% (appendix, table AIV) shows a decline as forecasts beyond 2012 are used for the EV/EBITDA multiple, but shows the expected annual increase for the EV/EBIT multiple, if only a small increase. So, the results of three measures contradict each other. The P/E multiple benefits most from the use of either 2013 or 2014 forecast data. The median of the absolute valuation errors and the percentage of valuation errors that is smaller than 15% (appendix, table AIV) suggest that the 2013 forecast data outperforms the 2014 forecasts, but the interquartile ranges suggest the use of 2014 forecast data leads to the most accurate value estimates.
19 answer on the second part of our research question, which proposes an increase in the accuracy of the value estimate as later-year forecasts are used instead of earlier-year forecasts for EV/EBITDA and EV/EBIT multiples. Our findings also are not in line with the findings in the literature concerning the use of later-year forward-looking P/E multiples in lieu of their earlier-year peers. Again, we examined the significance of these results with the use of the one-sided Mann-Whitney U test. The results of these tests are in brackets, in panel B of table V on page 17.
First, it becomes apparent that the use of any of the forecast years instead of the trailing data for the EV/EBITDA multiple never leads to a more accurate estimate of company value. This also implies that there is no difference in the valuation accuracy between the use of earlier and later-year forecasts. These results are negative probably because the EV/EBITDA multiple is not the best option when a lot of industries and companies of the same industry are pooled together. EBITDA excludes the depreciation expenditures that are required to achieve the profitability. This therefore makes it difficult to compare companies that own their residence or machine park versus companies that hire (or lease) their residence or machine park. Since, the multiple is estimated per industry, and after that no further distinction is made as all the industries are pooled together, the EV/EBITDA mulitple in this instance is probably not the best option. This same reasoning implies that the EV/EBIT mulitple is a better option for the non-financial industry subsample. As the three forward-looking EV/EBIT multiples are compared to their trailing peer, all three test results are significant. Thus, we reject these null hypotheses, and conclude that using any of the three forecast years in lieu of the trailing data leads to more accurate estimates of company value. However, as the difference between earlier and later-year multiples is examined, the null hypothesis is accepted all three times. So again, there is no difference in the accuracy of the value estimate as earlier-year forecasts are used instead of later-earlier-year forecasts.
The test on the P/E multiple shows no significant difference between absolute valuation errors of the 2011 trailing and 2012 forward-looking multiple. Nevertheless, the null hypotheses are rejected as the 2013 and 2014 forward-looking multiples are compared with the 2011 trailing multiple, which leads us to conclude that using these multiple leads to an increase in the accuracy of the value estimate. As the 2013 and 2014 forward-looking multiples are compared to the 2012 forward-looking multiple, the null hypotheses are also rejected. However, as there is no difference found between the use of the 2011 trailing and 2012 forward-looking multiple, this does not imply that the valuation accuracy further increases as later-year forecasts are used instead of earlier-year forecasts. This is also shown as we accept the null hypothesis, which states that the population locations of the distributions of 2013 and 2014 are equal.
20 does benefit from the use of forecasts instead of trailing data. Concerning the P/E multiple, our results again show that it benefits from the use of forecast data instead of trailing data, as was shown before in the literature (Kim and Ritter, 1999; Lie and Lie, 2002; Liu et al, 2002; Schreiner and Spremann, 2007). However, the results of the one-sided Mann-Whitney U test again do not support the second part of our research question, nor do they support the findings of Kim and Ritter (1999), Liu et al. (2002), and Schreiner and Spremann (2007) who established that the accuracy of the value estimate further increases as later forecast years are employed in lieu of earlier-year forecasts for the P/E multiple.
The financial companies (right-hand side in table V, page 17) should be easier to value, due to the liquidity of their assets (Lie and Lie, 2002), and are thus analysed separately from the other companies. The first thing that stands out is that the median of the absolute valuation errors of all three trailing multiples is larger than for the non-financial companies. Furthermore, their bias is also relatively large, but still only approximately 2% per multiple. However, the increase in accuracy as any forecast year is employed is also larger, and the bias decreases to being very small, and to being negligible. The median of the absolute valuation errors of the EV/EBITDA multiple decreases with 13.8%, that of the EV/EBIT multiple with 8.5%, and that of the P/E multiple with 8.1%, as 2012 forecast data is used in lieu of the trailing data. This increase in accuracy continues as 2013 forecasts are used instead of the 2012 forecasts, as the median of the absolute valuation errors decreases with 3.7%, 6.0% and 9.0%, respectively. Lastly, the use of 2014 forecast data leads to an improvement of 0.8% for the EV/EBITDA multiple, but it decreases the accuracy of both the EV/EBIT and P/E multiple with 0.9% and 2.2%.
21 with the findings in the literature (Kim and Ritter, 1999; Lie and Lie, 2002; Liu et al, 2002; Schreiner and Spremann, 2007).
Much like our earlier results, the null hypotheses comparing earlier to later-year forecasts are all accepted. This shows that there is no improvement in the accuracy of the value estimate as later forecast years are used instead of earlier years, for any of the three multiples. So again, we are unable to find significant results to positively answer the second part of our research question, and to support the findings of the literature that show that the accuracy of the multiples valuation method benefits from the use of later-year multiples in lieu of earlier-year multiples (Kim and Ritter, 1999; Liu et al, 2002; Schreiner and Spremann, 2007).
In summation, the main research question of this paper is confirmed by the one-sided Mann-Whitney U tests performed on the total sample and the country subsamples. The results show that the accuracy of the multiples valuation method indeed improves as any of the three forecast years is used instead of trailing data, for both multiples. However, as the analysis is performed on the industry subsamples, the results are less convincing. The one-sided Mann-Whitney U test results of the non-industry subsample show that, for the EV/EBITDA multiple, there is no significant increase in the accuracy of the value estimate as any of the three forecast years is used in lieu of the trailing data. For the financial companies, however, this increase in accuracy is significant for all three years. For the EV/EBIT multiple we find significant Mann-Whitney U test results for both industry subsamples. For both samples it proves to be worthwhile to use any of the three forecast years instead of the trailing data.
The second part of the research question, which states that using later-year forecasts instead of earlier-year forecasts for EV/EBITDA and EV/EBIT multiples results in a further increase of the accuracy of the value estimate, is not supported by our results. For the total sample and the country subsamples, the median of the absolute valuation errors and the other two measures do show a decrease as later-year forecasts are used. However, as we apply the one-sided Mann-Whitney U test to examine the significance of these differences, it shows that the later-year forward-looking multiples never produce significantly more smaller valuation errors. As the same analyses are applied to the industry subsamples, the results of the Mann-Whitney U test once again are insignificant, for both multiples.
22 subsamples, and the industry subsamples. But, more importantly, the difference in the accuracy of the value estimate is proven to be significant with the use of the one-sided Mann-Whitney U test. The results for the total and country sample show that using forecast data always results in significantly more smaller valuation errors. For the financial companies’ subsample, it also proves beneficial to use any of the three forecast years, and for the non-financial subsample there proves to be a significant difference for two of the three forecast years.
As we examine whether the accuracy of value estimate made by the multiples valuation method using the P/E multiple further increases as later forecast years are used in lieu of earlier forecast years, we again find that the median of the absolute valuation errors and the other two measures show an increase in accuracy for the total sample, the country subsample, and the industry subsample. This increase however, is smaller than for the difference between trailing and forward-looking multiples. As the one-sided Mann-Whitney U test is employed, we found only a few significant results. We found support for one out of three of our hypotheses with regard to the total sample, and we found support for one of the nine hypotheses with regard to the country subsamples. For neither of the industry subsamples, the Mann-Whitney U test finds a significant difference between the valuation errors produced by later and earlier-year forecasts. All in all, we found some evidence in support of the findings of Kim and Ritter (1999), Liu et al (2002), and Schreiner and Spremann (2007), but the evidence is much less convincing here, than it is for the first part of the research question.
We expected that employing the industry subsample would result in smaller valuation errors. This is indeed observed, as measured by the median of the absolute valuation error and the other two measures. However, the relative performance does not change for the EV/EBITDA and EV/EBIT multiple. For the P/E multiple, the industry results show that there is no benefit from using later-year multiples in lieu of earlier year multiples, whereas we did find some significant results on the total sample and country subsample. Overall,
VI. Conclusion
23 already established that the accuracy of the value estimated by the multiples valuation method using the P/E multiple benefits from the use of forecasts of earnings (Kim and Ritter, 1999; Lie and Lie, 2002; Liu et al, 2002; Schreiner and Spremann, 2007), and that this accuracy increases further as later-year forecasts are used instead of earlier-year forecasts (Kim and Ritter, 1999; Liu et al, 2002; Schreiner and Spremann, 2007). Despite those findings, the use of forecasts of earnings for the P/E multiple is also re-examined, as we apply a different method of analysis that is not used before. This method, the one-sided Mann-Whitney U test, allows us to determine the statistical significance of the difference of the valuation errors between the use of trailing and forecasts of earnings.
The analyses are performed on a sample of 1,388 companies, from Germany, France, and Great-Britain. The enterprise value and market capitalization are of ultimo 2011, with the trailing earnings data for calculation of the trailing multiples also from 2011. Median forecasts of EBITDA, EBIT, and net earnings are collected from 2012, 2013, and 2014, and make up the three forward-looking multiples. First, the analysis is performed on the entire sample, and on the separate countries. After that, further restrictions are imposed on the sample, and the analysis is performed on two industry subsamples, a financial and a non-financial companies sample. Estimating the multiples from industries increases the accuracy of the value estimate. Financial companies are separately analysed because they are easier to value, due to the nature of their assets.
24 smaller valuation errors. As the one-sided Mann-Whitney U test is a formal test, we consider it decisive for answering our research questions.
As hypothesized, the results of the analyses show that using forecasts instead of trailing data for EV/EBITDA and EV/EBIT multiples leads to a more accurate estimate of company value. The median of the absolute valuation errors of all three forward-looking multiples are smaller than the median of the valuation error of the trailing multiple, for both the for EV/EBITDA and EV/EBIT multiple, and similar patterns are shown by both the interquartile range and the percentage of absolute valuation errors that is smaller than 15%. The results of the one-sided Mann-Whitney U test support these findings. For the EV/EBITDA the results of the total sample, country subsamples, and the financial subsample show that the use of any of the forward-looking multiples provides a significantly higher amount of smaller valuation errors than using the trailing multiple does. However, all three null hypotheses are accepted as companies in the non-financial subsample are examined. This implies that there is no improvement as forecast data from any of the three years is used instead of trailing data. The results of the EV/EBIT multiple show that the forward-looking multiples produce significantly more smaller valuation errors, for the total sample and all subsamples.
The results concerning the second part of the research question are less conclusive. The decline in the median of the valuation errors is much smaller as we move from the use earlier-year to later-year forecasts. A similar small improvement in the accuracy of the value estimate is displayed by both the interquartile range and the percentage of absolute valuation errors that is smaller than 15%. Upon examining the increase in the accuracy of the valuation errors with the one-sided Mann-Whitney U test, the multiples valuation method never proves to benefit from the use of later-year forecasts instead of earlier-year forecasts, for either the EV/EBITDA or EV/EBIT multiple. We therefore conclude that using later-year forecasts instead of earlier-year forecasts for these enterprise value multiples does not result in more accurate estimates of company value.
25 absolute valuation errors that is smaller than 15% do show that there is an increase in the accuracy of the value estimate as later-year forecasts are used instead of earlier-year forecasts, however this is not as big as the difference between the trailing and each of the forward-looking multiples. In line with that, the results of the one-sided Mann-Whitney U show only two significant test results. One for the total sample, and one for the French subsample. The simple measures show results similar to the literature (Kim and Ritter, 1999; Lie and Lie, 2002; Schreiner and Spremann, 2007), but as the additional accuracy is examined by means of the one-sided Mann-Whitney U test, the difference proves to be is more often than not insignificant at a 95% confidence level.
Our results support the proposition that using forecasts of earnings in lieu of trailing earnings results in more accurate estimates of target company value. This is in line with valuation theory that argues that forecasted earnings are a better reflection of the long-term cash flow of a company (Koller et al., 2010), and that forecasts of earnings are based on more value relevant information (Damodaran, 2002). Nevertheless, the results do not show that the accuracy of the company value estimate increases further as later-year forecasts are used instead of earlier-year forecasts. There are some differences between the accuracy of the three forecasts years, but these become apparent as each of the three forecast years show different increases in accuracy as they are compared to the trailing data. This implies that practitioners should not simply choose the latest available forecast year, but rather choose the forecasted earnings that are the best representation of the long-term prospects of the business (Koller et al., 2010).
26 Furthermore, it will allow for a more thorough analysis of the origins of the differences in performance of the different multiples.
27 REFERENCES
Alford, A.A., 1992, “The Effect of the Set of Comparable Firms on the Accuracy of the Price-Earnings Valuation Method”, Journal of Accounting Research, 30 (1), 94 – 108.
Cooper, I., and Cordeiro, L., 2008, “Optimal Equity Valuation Using Multiple: The Number of Comparable Firms”, Working paper, London Business School.
Damodaran, A., 2002, “Investment Valuation”. Second edition, John Wiley & Sons.
Herrmann, V., and Richter, F., 2003, “Pricing with Performance-Controlled Multiples”, Schmalenbach Business Review, 55 (6), 194 – 219.
Kim, M., and Ritter, J.R., 2003, “Valuing IPOs”, Journal of Financial Economics, 53 (3), 409 - 437 Koller, T., Goedhart, M., and Wessels, D., 2010, “Valuation, Measuring and Managing the Value of Companies”. Fifth edition, University edition. John Wiley and Sons Inc.
Lie, E., and Lie, H.J., 2002, “Multiples Used to Estimate Corporate Value”, Financial Analysts Journal, 58 (2), 44 – 54.
Liu, J., Nissim, D., and Thomas, J., 2002, “International Equity valuation using multiples”, Journal of Accounting Research, 40, 135-172
28 APPENDICES
Table AI. Descriptive statistics of value drivers
Descriptive statistics of company value drivers. Numbers should be multiplied by € 1,000,000.
Mean Median Q25 Q75 N EBITDA 2011 219 22 6 122 1384 2012 332 50 14 240 870 2013 348 53 16 247 952 2014 408 72 22 310 871 EBIT 2011 152 16 4 84 1388 2012 233 35 9 163 893 2103 250 38 11 166 975 2014 298 54 17 217 890 Net earnings 2011 95 10 3 51 1388 2012 142 22 6 100 861 2013 157 25 8 109 936 2014 193 37 11 142 856
Table A II. Jarque-Bera test statistics from industry subsamples
A. Jarque-Bera test for non-normality, on the pooled industry distributions containing the absolute valuation
errors. Significant results prove non-normality, and are marked with an (*).
EV/ EBITDA EV/ EBIT P/E
29
Table AIII. Interquartile ranges and percentage of errors <15% of the total sample and country
subsamples
This table depicts the interquartile range of the absolute valuation errors, and the percentage of valuation errors that is smaller than 15%, of the total sample.
Interquartile range Percentage of valuation errors smaller than 15%
EV/EBITDA EV/EBIT P/E EV/EBITDA EV/EBIT P/E
30
Table AIV. Interquartile ranges and percentage of errors <15% of the industry subsamples
This table depicts the interquartile range of the absolute valuation errors, and the percentage of valuation errors that is smaller than 15%, for the industry subsamples.
Interquartile range Percentage of valuation errors <15%
EV/EBITDA EV/EBIT P/E EV/EBITDA EV/EBIT P/E
