The impact of past share prices on offer prices and deal success in
mergers and acquisitions
Koen Meurs1 June 2016
Abstract
The impact of past peak prices on offer prices is greater for targets that are listed in the United States than for targets that are listed outside the United States. Based on the prospect theory, I expect that shareholders use past peak prices as reference points in the evaluation of offer prices. An offer price that exceeds the past peak price would then result in a jump in deal success. My empirical findings support this view for companies that are listed in the United States. I do not find proof for this view in companies that are listed outside the United States. A potential explanation for the difference in findings is the widely dispersed shareholding in the United States.
Field key words: Mergers, Acquisitions, Offer price, Reference point, Behavioural corporate finance JEL classification: G31, G34
This thesis is submitted as a master’s thesis for the Master of Science in Finance (EBM866B20). Supervisor: Dr. ing. N. Brunia, University of Groningen. I would like to thank Dr. ing. Brunia for his useful feedback.
2
1. Introduction
Capital allocation is one of the most important decisions that the management of a company faces. The management can decide to pay down debt, pay dividends or buyback stock. Another opportunity is to participate in M&A transactions (KPMG, 2015).
After the management has identified a potential takeover target, they have to assess what they want to offer for the shares of the firm. They have to make various assumptions about e.g. the potential synergies and the bargaining power of the target which makes the valuation complicated.
A high offer price will increase the likelihood that the offer is accepted but decreases the subsequent return on investment.
In this thesis I analyse whether historical share prices of the target can guidance the management of the bidder to a satisfactory offer price. Baker, Pan, and Wurgler (2012) suggest that bidders
anticipate on the use of past peak prices as reference prices by the shareholders of the target. As a result, the bidders offer a higher premium when the past peak price of the target is higher. I analyse whether the managements of the bidders use past peak prices of the target’s shares to determine the offer price, and whether the shareholders of the target do indeed use past peak prices in the assessment of offer prices.
The shareholders of the target will be aware of past peak prices as these peaks are recently achieved and frequently published in the media (Baker et al., 2012). Tversky and Kahneman (1974) show that people stay too close to an initial value in their assessment of a value. In an M&A context, this would result in past peak prices that play a large role in the assessment of the offer price by the
shareholders. Shefrin and Statman (1985) show in an empirical research that investors are reluctant to realize losses. This reluctance should result in a higher rate of deal success for offers that exceed the past peak price.
When past peak prices act as reference points for the target’s shareholders, it is efficient for bidders to set the offer price slightly higher than one of the past peak prices. When the offer price exceeds the reference point, it will result in a jump in deal success.
My sample consists of 11,431 takeover bids that are announced in the years 2000 to 2015. I do not apply geographical restrictions.
3
find that bidders are more prone to use past peak prices for targets that are listed in the United States.
My results support the view that shareholders of target companies that are listed in the United States use past peak prices to assess the height of the offer price. I find a significant jump in deal success when past peak prices are exceeded.
I do not find evidence for the use of past peak prices as reference points by shareholders of targets that are listed outside the United States. A potential explanation for the difference in findings is that the shareholding is widely dispersed in the United States (Franks, Mayer, and Rossi, 2009).
The management of bidding firms should consider past peak prices in the determination of the offer prices, but only when the target firm is listed in the United States.
My research adds to the current literature in multiple ways.
First, I use a worldwide sample which enables the comparison between regions. Previous related research focused on a specific country or continent. Secondly, I perform statistical tests to determine the breakpoints in the relation between the past peak price and the offer price. This is in contrary to previous research that used ‘visual inspection’ to determine the breakpoints. Lastly, I compare the role that peak prices of different periods play in the determination and reception of offer prices. Previous research focussed on the 12-month peak price.
Section 2 provides a theoretical background and the hypothesis that are tested in this thesis. Section 3 explains the methodology used and describes the data. Section 4 presents the results and section 5 concludes.
2. Literature
When a company intends to acquire a firm, they have to convince the targeted shareholders to sell their shares. Crucial in this is the “offer price” which the bidder offers on the shares of the target. In a merger, the management of the bidder and the target come to an agreement on the price per share. Afterwards, the shareholders of the target have to vote against or in favour of the proposal
(Offenberg and Pirinsky, 2014).
4
On average, the offer price includes a takeover premium which ranges between 20 and 30% of the target firm’s share price prior to the acquisition (Petitt and Kenneth, 2013).
For tender offers, the takeover premium is in general higher than for mergers (e.g. Offenberg and Pirinsky, 2014).
A widely mentioned rationale for the payment of a takeover premium are the estimated synergies (Hillier, Grinblatt, and Titman, 2011) which can be reaped when the target is acquired. Synergies increase the value of the company when the present value (PV) of the free cash flows (FCFs) of the combined company outreach the sum of the PVs of the separate FCFs. An example of synergies is the reduction in costs when employees can be laid off after the merge of two firms.
Another explanation for the payment of takeover premiums is the hubris theory of Roll (1986). This theory describes the value of a company as a random variable with a mean equal to the current market valuation. The bidders know that the target’s shareholders will only sell their shares when the offer price exceeds the current market price. When the random variable (the value of the company) does not exceed the market price, the bid is abandoned. Therefore, all observed offers are in the right tail of the distribution of valuations.
Bidders that suffer from hubris are convinced that their valuation of the potential target and the value of the estimated synergies of the potential target are correct. When the random variable exceeds the current market value of the target, they will make an offer on the target. They abandon the offer when the random variable is lower than the current market value of the target. The takeover premium is the result of this bias in observed offers (Roll, 1986).
Rational bidders realize that valuations are prone to errors and that valuations with negative errors are not visible. As a result, rational bidders will adjust their offer adequately to incorporate the potential valuation errors (Roll, 1986).
Baker et al. (2012) introduce a new theory that determines the height of the paid premiums. Their theory is based on the fact that valuing a company is subjective. They state that this subjectivity makes the offer price and its reception feasible to include psychological influences on the bidder and the board and shareholders of the target.
5
evaluation of the offer and that the managements of bidding companies use past peak prices in the determination of the offer.
The theory behind the use of reference prices in M&A is based upon two theories, the prospect theory of Kahneman and Tversky (1979) and the anchoring theory of Tversky and Kahneman (1974).
Rationale for the use of reference points
There are various situations in which people have to make an estimate on a certain value. In the process to achieve this estimate, people often have an initial value which is the result of e.g. a basic computation. Tversky and Kahneman (1974) show in their paper that people stay too close to this initial value and do not sufficiently include new data and other considerations in the assessment of the value. This insufficient adjustment is called anchoring.
The determination and reception of the offer price can be prone to anchoring for two reasons. First, when a company receives a takeover offer, the shareholders have to assess what the company is worth in their belief. The past peak of the stock prices are widely published in the media and can therefore act as relevant reference / anchor points (Baker et al, 2012). Shareholders will be more likely to accept an offer when it exceeds their reference price.
Secondly, the management of the target will pursue the highest possible price. Relevant or not, past peak prices can be used as anchoring points during the negotiations. The management of the bidder can use the past peak price to indicate that the target is worth the offer price, since the company was valued at that level in the past, even without the synergy that the bidder estimates to realize. The second theory is the prospect theory that Kahneman and Tversky published in 1979. At that time, the expected utility theory was the generally accepted theory to study decisions under risk. The expected utility theory has proven to be very successful in the description of how people should behave. Kahneman and Tversky (1979) publish empirical evidence which shows that the expected utility theory does not fully account for observed behaviour in decision-making under risk.
6
Figure 1 presents the shape of the value function. This shape is based on the observed decision taking under risk. People are risk averse for gains, which results in a concave relationship. For losses, people are risk seeking. This risk seeking behaviour results in a convex relationship in the domain where the offer price is lower than the reference price. The curve is steeper in the domain of the losses since people are in general risk averse.
Hypothesis
Baker et al. (2012) suggest that bidders who anticipate on the use of past peak prices as reference prices among investors, offer a higher premium when the past peak price of the target is higher. This line of reasoning leads to a positive relation between the past peak price and the offer premium, hence hypothesis 1:
𝐻1 : There is a positive relation between the offer premium and the past peak price of the target.
Baker et al. (2012) analyse the effect of the 52-week high reference point on the offer price in a sample of 7,020 deals between January 1, 1984 and December 31, 2007. They require the target to be listed in the United States. They find a positive relation between the past peak price and the offer premium.
Niinivaara (2010) bases his research on the working paper of Baker, Pan, and Wurgler (2009). He investigates the relation between the offer premium and the 52-week high in a European sample. In his sample of 3,009 offers, he finds a significant and positive relation between the offer premium and the 52-week high.
7
Niinivaara (2010) interviews investment bankers to underpin the results with practical experience. The investment bankers confirm that past price levels are used as negotiation arguments and that during the determination of the offer price psychology can be a consideration.
According to the prospect theory, the marginal perceived loss is lower when the current price is further from the past peak price. This can result in the shareholders of the target that accept an offer relatively easy when the offer price is relatively far from the past peak price. Additionally, there can be valid reasons (e.g. financial distress) why the share price of the target dropped since the past peak price. Therefore, I expect that the impact of past peak price on the offer is greater when the past peak price is relatively low.
This leads to the following hypotheses:
𝐻2 : The impact of past peak prices on the offer price is greater for peak prices that are relatively low.
Baker et al. (2012) and Niinivaara (2010) test this hypothesis and find indeed that the relation is steeper for past peak prices that are closer to the share price 30 days prior to the announcement date.
The use of reference points among the shareholders of the target should be visible in the analyses of deal success. Shefrin and Statman (1985) showed that investors are reluctant to realize losses. This reluctance should result in a discontinuous jump in deal success when the reference price of shareholders is exceeded.
A positive relation between the offer price and deal success can be explained by the expected utility theory as well. A higher offer price leads to higher utility for the shareholders of the target. However, when there is a discontinuous jump in deal success when the reference price is exceeded, this cannot be explained by the expected utility theory. When the past peak price is exceeded, the shareholders of the target encounter an increase in perceived value. This will result in a discontinuous jump in deal success.
8
I set the following hypothesis to test the influence of a surpass of the past peak price on deal success:
H3 : Deal success increases discontinuously when the offer price meets or exceeds the targets past peak price.
Baker et al. (2012) and Niinivaara (2010) find a discontinuous jump in deal success of 4 to 6% when the (52-week high) reference price is exceeded. This finding suggests that the shareholders of the target use reference points in the assessment of the height of the offer price.
Period to determine peak price
The use of the 52- week high to test for the existence of reference points is custom. Chira et al. (2015) describe various researches (e.g. when investors sell stock, trading volume, IPO activity) in which the 52-week high is used as a reference point by investors.
Baker et al. (2012) investigate the influence of the past peak prices from 13 weeks to 104 weeks (in steps of 13 weeks). They find that it is common for bidders to offer exactly a past peak price. For simplicity reasons, they decide to focus on the 52-week past peak price in their research. Niinivaara (2010) focuses on the 52- week peak price. Additionally, he runs a linear regression without control variables on the 1 to 15 month peak prices. He finds that the 1-month peak price has the highest impact on the offer premium.
Research of Baker et al. (2012) and Niinivaara (2010) shows that the 52-week peak price is not by definition the ideal term in this context. Therefore, I compare the results for the 2 to 24 months peak prices to see which past peak price is most relevant. Since the use of the 52-week peak price is most common, I test the following hypotheses:
𝐻4𝑎 : The impact of the 52-week peak price on the offer premium is greater than the effect of the
other past peak prices.
𝐻4𝑏: The impact of the 52-week high on deal success is greater than the effect of the other past peak
prices.
The influence of shareholder concentration
9
investors are likely to have more resources available to assess the height of the offer price. I test this line of reasoning with the following two hypotheses:
𝐻5𝑎: Past peak prices have a greater influence on the offer price for companies that are listed in the
United States.
𝐻5𝑏: The discontinuous increase in deal success when the offer price meets or exceeds the targets
peak price is larger when the target is listed in the United States. Control variables
Table 1 Expected signs control variables
The expected signs for the control variables in the regressions based on the literature.
Control variable Expected sign: offer premium Expected sign: Deal success
Payment in shares - - Payment in cash + + Target Book-to-market + / - n.a. Market capitalization - - Volatility + n.a. Bidder Book-to-market + n.a. Market capitalization - / + +
2.5.1 Control variables offer premium Deal characteristics
I expect a positive coefficient for the dummy variable “payment in cash” and a negative coefficient for the dummy variable “payment in shares” for two reasons. First, de La Bruslerie (2013) shows that there is a trade-off for the target; the target can either receive a high offer premium and be paid in cash, or the target can benefit from potential future returns and be paid in shares. Secondly, shareholders are immediately exposed to capital gain taxes when the offer premium is paid in cash. The payment of taxes can be deferred when the offer premium is paid in shares.
Target: Book to market (+/-)
10
Georgen and Renneboog (2004) find a negative relation between the BM and the offer premium. Comment (1995), Niinivaara (2010) and Schwert (2000) do not find a significant coefficient for the BM of the target.
Target: Market cap (-)
De La Bruslerie (2013) mentions that inside owners or large block holders trade their private benefits for a higher premium. For larger targets, the premium can be spread over a larger investment for the bidder. Alexandridis, Fuller, and Terhaar (2013) show that the offer premium is lower for large targets. This finding is in line with increased complexity related to the acquisition of larger targets which makes it harder to achieve the assumed economic benefits. Additionally, larger firms are in general better managed which reduces the potential economic benefits (Simonyan, 2014). These arguments lead to a negative sign for the target market capitalization.
Target: Volatility (+)
Simonyan (2014) proposes that the market volatility is related to the offer premium through investor sentiment. He finds that takeover premiums are higher when investors are bearish and that the market volatility increases when investors become more bearish. I expect a positive relation between the share price volatility and the offer premiums.
Bidder: Book to market (+)
Bidders with high growth opportunities (low BM) pay significantly higher offer premiums (Officer, 2003). Therefore, I expect a positive sign for the BM ratio. Baker et al. (2012) do not find a significant result for the BM ratio of the bidder.
Bidder: Market cap (+/-)
11
2.5.2 Control variables deal success
Deal characteristics
The literature on deal success suggests that cash deals are more likely to be accepted. A reason for this is that stock swap offers need approval from both the shareholders of the target and the shareholders of the bidder. Offers in cash require only approval from the shareholders of the target (Branch, Wang, and Yang, 2008).
Market cap
Empirical results of Baker et al. (2012), Schwert (2000), and Officer (2003) show that offers on large targets are less likely to be accepted, while offers by large bidders are more likely to be completed. As a result, I expect a negative sign for the market cap of the target and a positive sign for the market cap of the bidder.
3. Data and Methodology
In this section I describe the data and methods I use to test the hypotheses introduced in the previous section.
Table 2 Selection criteria for the dataset
All the offers are from Zephyr. For offers to be included in the results from Zephyr, they had to fulfil the following requirements: (i) The offer price is not missing and (ii) the intention of the offer is acquisition, institutional buy-out, MBI, MBO or merger and (iii) the bidder starts with less than 50% of the target firm shares outstanding. I set further restrictions to exclude offers with missing target ISIN numbers or announcement dates. I exclude offers for which the acquirer sought less than 50% of the shares outstanding and offers for which the targets do not have share prices in DataStream. Lastly, I exclude offers with an announcement date (in Zephyr) prior to 01/01/2000 or after 31/12/2015.
Zephyr result: 16,165
Missing Target ISIN: 2,521 Missing Announcement date 685 Desired acquisition <50% 823 Share prices not available 448 Announcement prior to 2000 209
Announcement in 2016 48
Final sample 11,431
Sample selection
12 A. The offer price is not missing;
B. The announcement date of the offer is not missing; C. The target has an ISIN number;
D. The share price in the period 730 days before the offer date to (and including) the offer date is not missing;
E. Zephyr indicates the deal type as a(n) acquisition, institutional buy-out, management buy-in, management buyout or merger;
F. The bidder starts with less than 50% of the target firm shares outstanding; G. The sought stake is at least 50%;
H. The announcement date is between the 01-01-2000 and 31-12-2015.
Criterion A, B and D enable the calculation of the offer premiums and peak prices. Criterion C is required to be able to retrieve the right share price information.
Criterion E is in line with Baker et al. (2012) and Niinivaara (2010). The criterion is set to exclude recapitalizations, share repurchases and self-tendered offers as these offers do not result from a traditional negotiation process.
Criteria F and G are set to include only offers which are meant to result in a change in control of the target. Following Baker et al. (2012) and Niinivaara (2010), I require companies to make an offer for at least 50% of the shares. Therefore, I selected deals for which the target had an initial stake <=50%, and the sought stake was >50%.
I set criterion H since the Zephyr database is incomplete prior to 2000 (DSE Unibo, 2016). Therefore, to remain a complete dataset in the years under investigation, I start the dataset at 01-01-2000. I end the sample at 31/12/2015 to cover only full years.
There are no geographical restrictions. The Zephyr database contains worldwide M&A activity.
Data construction
The research concerns the relation between the offer price and past peak prices. In the literature, these prices are not applied in raw form but transformed to values relative to the share price 30 days prior to the announcement date of the offer.
Following Baker et al. (2012), I calculate the offer premium that is bid on target i as
𝑃𝑅𝐸𝑀𝑖 =LN ( 𝑂𝑖
13
where 𝑂𝑖 is the offer price that is bid on a share of target i and 𝑃𝑖,𝑡−30 is target i’s share price 30
calendar days prior to the announcement. When 𝑡−30 is not a trading day, I take the share price of
the last trading day prior to 𝑡−30.
I scale the values to overcome the problem with heteroscedasticity that would arise if I compared the offer prices in raw form. I apply a lagged scaling factor to exclude any new information or rumours that would influence the offer price (Baker et al., 2012).
To test the hypotheses, the highest share price achieved in the past period is required. I calculate this highest share price ph as
𝑝ℎ𝑖,𝑥= max (𝑃𝑖,𝑡−31, 𝑃𝑖,𝑡−32, … … , 𝑃𝑖,𝑡−𝑥 ) (2)
where 𝑝ℎ𝑖,𝑥 is the highest share price of target i achieved between day t-31 and day t-x, x is an
integer in the range 31 to 730 and t is the announcement day of the offer.
I calculate the peak price for the past 2 to 24 months, excluding the 30 days prior to the
announcement. I use months with 30 calendar days. Months 12 and 24 are an exception with 35 calendar days to come to 365 calendar days in a year. I do not calculate the 1-month peak price since this value is subject to pre-announcement run-up (Baker et al., 2012). The formula for the 6-month peak price of company i is displayed below:
𝑝ℎ𝑖,180= max (𝑃𝑖,𝑡−31, 𝑃𝑖,𝑡−32, … … , 𝑃𝑖,𝑡−180 ) (3)
14
Figure 2 Time frame to determine past peak price.
The time frame to determine the past peak price of target i. Days t up to t-30 are excluded in the determination of the peak price. x is the first day included in the determination of the peak price. Day t-30 acts as a scaling factor for the peak price and offer premium.
I follow Baker et al. (2012) and scale the peak price of the target by the target’s share price 30 calendar days prior to the announcement, leading to a relative value of the peak price 𝑃𝐻𝑖,𝑥
𝑃𝐻𝑖,𝑥 =LN ( 𝑝ℎ𝑖,𝑥
𝑃𝑖,𝑡−30 ) (4)
where 𝑝ℎ𝑖,𝑥 is the highest share price of target i achieved between day t-31 and t-x prior to the
announcement bid.
I winsorize the offer premiums and peak prices at the 5th and 95th percentile. Winsorizing is a common procedure to achieve robust statistics (Ghosh and Vogt, 2012). Any data above the 95th percentile is replaced with the value at the 95th percentile, and any value below the 5th percentile is replaced with the value at the 5th percentile. Baker et al. (2012) winsorize at the 99% level.
Niinivaara (2010) winsorizes the offer premium at 5% and the other variables at 1%. My robustness checks indicate that my results are robust to the winsorization (see appendix A.7 and A.8)
Sample description
15
Table 3 Offer premium and past peak prices.
The offer premium is the offer price from Zephyr expressed as a log percentage difference from the share price 30-days prior to the announcement date from DataStream. The X- month high price is the highest share price between day t-31 and t-x, expressed as a log percentage difference from the share price 30-days prior to the announcement date.
Mean Standard deviation 5%-percentile Median 95%- percentile
Offer premium % 24.30 27.74 -28.12 20.91 90.16
2 - month peak price % 6.54 8.22 -0.29 3.37 29.63
3 - month peak price % 10.06 12.14 0.00 5.41 43.98
4 - month peak price % 12.94 15.57 0.00 7.01 56.61
5 - month peak price % 15.70 19.07 0.00 8.20 69.31
6 - month peak price % 17.81 21.46 0.00 9.13 77.63
7 - month peak price % 20.53 24.80 0.00 10.29 89.61
8 - month peak price % 22.86 27.72 0.00 11.54 100.95
9 - month peak price % 25.09 30.34 0.00 12.52 109.86
10 - month peak price % 27.14 32.74 0.00 13.55 117.87
11 - month peak price % 29.14 35.05 0.00 14.76 125.77
12 - month peak price % 31.58 38.07 0.00 16.02 137.48
13 - month peak price % 33.34 39.97 0.00 16.85 143.51
14 - month peak price % 35.06 41.72 0.00 17.95 149.17
15 - month peak price % 36.84 43.72 0.00 18.83 156.50
16 - month peak price % 38.46 45.44 0.00 19.86 162.00
17 - month peak price % 40.03 47.25 0.00 20.63 167.90
18 - month peak price % 41.63 49.12 0.00 21.51 174.70
19 - month peak price % 43.26 50.72 0.00 22.31 179.18
20 - month peak price % 44.83 52.58 0.00 23.17 186.63
21 - month peak price % 46.23 54.07 0.00 24.23 191.51
22 - month peak price % 47.63 55.54 0.00 25.13 196.44
23 - month peak price % 49.06 57.08 0.00 26.10 201.49
24 - month peak price % 50.43 58.54 0.00 26.77 206.55
The zeros at the 5th percentile represent offers for which the peak price is equal to the share price 30 days prior to the announcement dates. Day 𝑃𝑖,𝑡−30 is not included in the period to determine the
past peak price. When 𝑡−30 is a non-trading day, the share price in DataStream is equal to 𝑃𝑖,𝑡−31
which is included in the sample.
16 0 5 10 15 20 25 30 35 0 200 400 600 800 1.000 1.200 1.400
N = offers Mean offer premium %
Figure 3: Number of offers and average offer premium over time.
They only include offers in the US, where the drop in M&A activity was more visible than in the rest of the world. I see a drop in the number of offers in the US in 2008 as well. The average offer exceeds the share price 30 days prior to the announcement date by 20 to 30%.
.
Control variables
I follow Baker et al. (2012) and Niinivaara (2010) in the selection of the control variables. Due to lack of data, I am unable to include dummy variables regarding the attitude of the deal (hostile or not), whether the deal is a tender offer or not and whether the offer is made by a financial buyer.
Table 4 – Included control variables
The table shows which control variables are included in the research. Payment in cash and Payment in shares are dummy variables that are coded 1 when that type of payment is included in the offer. I calculate the Book-to-Market as 1/MTBV from DataStream. Variable: Included for Target Included for Bidder
Abbreviation: Source: DataStream Code:
Dummy variable payment in cash n/a n/a CASH Zephyr n/a
Dummy variable payment in shares n/a n/a SHARES Zephyr n/a
Book-to-Market Yes Yes BM DataStream MTBV
Return on Assets Yes Yes ROA DataStream WC08326
Share price volatility Yes No VOL Own
calculations
n/a
Market capitalization Yes Yes MCAP DataStream MV
Table 4 shows which control variables are included in the research.
17
The Book-to-Market ratio is the balance sheet value of equity to market value of equity (BM). The Return on Assets (ROA) is defined as the net income before preferred dividends and tax over total assets. In line with Baker et al. (2012), I calculate the share price volatility (VOL) as the standard deviation of the return of the target on a daily base, for the year prior to the announcement date. Following Baker et al. (2012), I use the 10log of the market capitalization (MCAP) from DataStream. For the BM, ROA and MCAP, I take the values at the year-end prior to the announcement date from DataStream. I express all control variables, except MCAP and the dummy’s for deal characteristics, in percentages. The BM, ROA, VOL and MCAP are all winsorized at the 5th and 95th percentile.
Table 5 provides an overview of the mean, standard deviation, 5th percentile, median and 95th percentile of the control variables.
For only 2,447 offers all control variables are available. I compare the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) to determine which control variables can be removed from the model without reducing the power. AIC and BIC measure the fit and complexity of a model. The fit of the model is expressed in negative values, the complexity of the model is
expressed in positive values. A better fit for the level of complexity results in lower values for the AIC and BIC. The difference between the AIC and BIC is that the BIC includes the natural logarithm of the number of observations in the formula (StataCorp. 2015)
Table 5 Control variables
The BM, ROA and MCAP are taken at the year-end prior to the announcement date. All continues variables are Winsorized at the 5th and 95th percentile.
N Mean Standard deviation 5%-percentile Median 95%- percentile Winsorized Dummy payment in Shares 11,431 0.26 0.44 0 0 1 No Dummy payment in Cash 11,431 0.64 0.48 0 1 1 No Target BM % 9,159 75.51 64.73 4.60 55.87 256.41 Yes
Target ROA % 5,569 -1.53 16.06 -48.11 2.62 18.80 Yes
Target MCAP 10,199 8.29 1.03 6.46 8.26 10.20 Yes
Target volatility % 11,431 3.67 2.07 1.30 3.05 8.88 Yes
Bidder BM % 5,528 63.57 44.18 10.95 51.28 175.44 Yes
Bidder ROA % 5,562 3.17 10.40 -26.89 4.55 18.56 Yes
18
Regressions and tests
To test hypothesis 1, I analyse the relation between the peak prices 𝑃𝐻𝑖,𝑥 and the offer premiums
𝑃𝑅𝐸𝑀𝑖 by the formula
𝑃𝑅𝐸𝑀𝑖= α+ β(𝑃𝐻𝑖,𝑥) + 𝜖 𝑖,𝑡 (5)
with Huber-White standard errors.
Huber white standard errors can deal with problems about normality, heteroscedasticity and observations that have large residuals, leverage or influence. The coefficients of the regression are unadjusted, the standard errors are adjusted for issues with heterogeneity and lack of normality (UCLA: Statistical Consulting Group, 2016).
In all regressions, I control for the inverse of the share price 30 days prior to the announcement. I add this control variable for the reason that I scale all variables by the 30-day lagged price, and the boards and shareholders of target and bidder might not think in scaled prices. The inclusion of this control variable avoids the measurement error leading to a spurious positive correlation that arises otherwise (Baker et al, 2012). The correlation of the 12-month peak price with the offer price is 0.2. This correlation rises to 0.65 after the scaling of the offer premiums and peak prices.
When hypothesis 2 is true, a linear regression is not suitable to describe the relation between the offer premiums 𝑃𝑅𝐸𝑀𝑖 and the peak prices 𝑃𝐻𝑖,𝜒. Following Baker et al. (2012) and Niinivaara
(2010), I test for a non-linear relationship.
Baker et al. (2012) and Niinivaara (2010) run a Gaussian kernel regression for the offer premium on the peak price. A kernel regression is a type of nonparametric regression. Nonparametric regressions do not make strong assumptions about the shape of the true regression function (Altman, 1992). Therefore, nonparametric regressions are helpful to check parametric models, for data description and when the shape of the curve is not known. The smoothness of the kernel regression is influenced by the bandwidth of the regression. The bandwidth determines the range of values of observations that are included to estimate the regression. A larger bandwidth results in a smoother shape of the regression function (Altman, 1992).
The Gaussian kernel regression is a specific form of the kernel regression that assumes that the error terms are normally distributed (Niinivaara, 2010).
19
confirm that the breakpoints derived from the Gaussian kernel regression are non-arbitrary. The ‘nl’ (non-linear) command in Stata uses iteration to determine the breakpoint for a regression based on least-squares (UCLA: Statistical Consulting Group, 2016). Stata requires initial values of the
coefficients and breakpoints to correctly iterate the coefficients and breakpoints. I use the
breakpoints determined by visual inspection of the Gaussian kernel regression as an initial starting point.
The non-linear regression is run as
𝑃𝑅𝐸𝑀𝑖,= (𝛼1+ 𝛽1(𝑃𝐻𝑖,𝑥)) ∗ (𝑃𝐻𝑖,𝑥 < ø) + (𝛼1+ 𝛽1∗ ø + 𝛽2∗ (𝑃𝐻𝑖,𝑥− ø)) ∗ (𝑃𝐻𝑖,𝑥≥ ø)
+ 𝜖 𝑖,𝑥 (6 )
in which ø is the breakpoint iterated to by Stata.
Additionally, I test whether there is a second breakpoint prior to ø.
𝑃𝑅𝐸𝑀𝑖,= 𝛼1+ 𝛽1(𝑃𝐻𝑖,𝑥) ∗ (𝑃𝐻𝑖,𝑥 < ¥) + 𝛼1+ 𝛽1∗ ¥ + 𝛽2(𝑃𝐻𝑖,𝑥− ¥) ∗ (𝑃𝐻𝑖,𝑥 ≥ ¥) ∗
(𝑃𝐻𝑖,𝑥 < ø) + 𝛼1+ 𝛽1∗ ¥ + 𝛽2∗ (ø − ¥) + 𝛽3∗ (𝑃𝐻𝑖,𝑥− ø) ∗ (𝑃𝐻𝑖,𝑥≥ ø)+ 𝜖 𝑖,𝑥 (7)
where ¥ is the breakpoint prior to ø and ø is the breakpoint found in the prior test. Or whether there is a breakpoint after ø with the formula
𝑃𝑅𝐸𝑀𝑖,= 𝛼1+ 𝛽1(𝑃𝐻𝑖,𝑥) ∗ (𝑃𝐻𝑖,𝑥 < ø) + 𝛼1+ 𝛽1∗ ø + 𝛽2(𝑃𝐻𝑖,𝑥− ø) ∗ (𝑃𝐻𝑖,𝑥 ≥ ø) ∗
(𝑃𝐻𝑖,𝑥 < 𝛬) + 𝛼1+ 𝛽1∗ ø + 𝛽2∗ (𝛬 − ø) + 𝛽3∗ (𝑃𝐻𝑖,𝑥− 𝛬) ∗ (𝑃𝐻𝑖,𝑥≥ 𝛬)+ 𝜖 𝑖,𝑥 (8)
where ø is the breakpoint found in the test with 1 breakpoint and 𝛬 is the potential breakpoint after ø.
Following Baker et al. (2012) and Niinivaara (2010), I perform piecewise linear regressions to cover the non-linearity in the relationship between the offer premium and the peak price. Piecewise linear regression has to be applied when the relationship between the dependent variable y and an independent variable x is different for different ranges of x. The piecewise linear regression allows the data to be described by multiple linear models which are split by a breakpoint (Ryan and Porth, 2007). I include the breakpoints that I estimate by the non-linear regression described above. I run this regression as
𝑃𝑅𝐸𝑀𝑖= α+ 𝛽1(𝑚𝑖𝑛(𝑃𝐻𝑖,𝑥, ø))+ 𝛽2𝑚𝑎𝑥(0, (𝑃𝐻𝑖,𝑥− ø)) + 𝜖 𝑖,𝑥 (9)
20 With two breakpoints, the regression is run as
𝑃𝑅𝐸𝑀𝑖= α+ 𝛽1(min (𝑃𝐻𝑖,𝑥, ¥))+ 𝛽2max (0, min( 𝑃𝐻𝑖,𝑥− ¥, (ø − ¥)))
+ 𝛽3𝑚𝑎𝑥(0, 𝑃𝐻𝑖,𝑥− ø)+ 𝜖 𝑖,𝑥 (10)
with Huber-White standard errors.
The piecewise linear regression allows for a marginal effect of 𝛽1 for premiums up to ¥ and 𝛽2 for the
part of the premium between ¥ and ø. 𝛽3 allows for the marginal effect of the part of the premium
beyond ø. Considering the S-shaped value function, we would expect a 𝛽1 > 𝛽2 > 𝛽3 , since the
marginal effect should decrease with higher peak prices.
Probit regression
To test hypothesis 3, I analyze the relation between the offer price and deal success. I follow Baker et al. (2012) and Niinivaara (2010) and run a probit regression on deal success. Probit regressions are used to model a binary outcome variable (UCLA: Statistical Consulting Group, 2016). I run the regression
Ω𝑖 = 𝛼 + 𝛽(𝜋𝑖), + 𝜖 𝑖 (11)
where Ω is a dummy variable coded 1 for completed offers and 0 for not completed offers. π is a dummy variable that is coded 1 when the offer price exceeds the peak price and 0 otherwise. The likelihood that an offer is accepted will most likely increase when the offer price is higher and exceeds the peak price. To test for a true discontinuity in the relation when the peak price is exceeded, I follow Baker et al. (2012) and Niinivaara (2010) and run additional tests with the
polynomial to the power 2, 3 and 4 included. When the relation between Ω𝑖 and 𝜋𝑖 is continuous, the
relation will be described by the polynomials. As a result, the coefficient of the dummy variable 𝜋𝑖
will no longer be significantly different from zero. A significant value for 𝜋𝑖 shows that there is a jump
in deal success when the peak price is exceeded.
4. Results
21
Table 6 – Linear regression of the offer premium on the past peak price
Regression of the offer premium on the 2, 6 and 12 month peak prices. Offer premium is the offer price from Zephyr expressed as a log percentage difference from the share price 30-days prior to the announcement date from DataStream. The x- month high is the highest share price of company i between day t-x and 31 days prior to the announcement date, expressed as a log percentage difference from the share price 30-days prior to the announcement date.
(1) (2) (3)
VARIABLES 2-month high 6-month high 12-month high β 0.500*** 0.258*** 0.145*** (0.0406) (0.0158) (0.00902) Inverse P 0.246*** 0.149** 0.125** (0.0597) (0.0588) (0.0591) Constant 20.66*** 19.48*** 19.54*** (0.320) (0.310) (0.306) Observations 11,431 11,431 11,431 R-squared 0.029 0.045 0.044 F 94.37 155.6 151.3
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
A 10% higher share price during the 2- months peak price time span results in a 5.00% higher offer premium. A 10% higher peak price set in the 6 months prior to the announcement, results in a 2.58% higher offer premium. For the widely tested 12-month high (52-week high), I find that a 10% higher peak price results in a 1.45% higher offer premium. This 1.45% is significantly lower (p=0.000) than for the peak prices that are set in the 2 to 11-month periods. Baker et al. (2012) find a beta of 0.96% and Niinivaara (2010) finds a beta of 0.83% in his sample. The beta’s they find are significantly lower (p=0.000) than the beta of 1.45% I find.
My model has an squared of 4.4% for the 12-month peak price. This is comparable to the R-squared of 4% Niinivaara (2010) finds and lower than the 8% Baker et al. (2012) find.
22
To test hypothesis 2, I have to test for a non-linear relation between the offer premium and the peak price. Following Baker et al. (2012) and Niinivaara (2010), I run a Gaussian kernel regression to determine the overall shape of the relation between the peak price and the offer premium.
Figure 5 presents the Gaussian kernel regression for the 12-month peak price. The figure shows that there is a positive relation between the offer premium and the peak price. In table 6 we found that this relation has a beta of 0.145. Based on visual inspection, I determine that either 18% or 28% is a good starting point for the non-linear regression function. Since I work with natural logarithms, a value of 18% (28%) can be interpreted as a peak price that exceeds 𝑃𝑡−30 by approximately 20%
(32%). Baker et al. (2012) use a breakpoint of 25%. Figure 4 Gaussian kernel regressions
Gaussian kernel regression of the relation between the offer premium and 12-month peak price. The regression displayed in the left figure has a bandwidth of 3. The regression displayed in the right figure has a bandwidth of 0.4. The bandwidth determines the range of observations that is included to determine a certain point in the regression. The larger the bandwidth, the smoother the regression line is.
I run two non-linear regressions in stata. I indicate an initial value for the breakpoint of 18% for the first regression and an initial value of 28% in the second regression.
Stata iterates to a breakpoint at 18.57171 when I use 18% as the initival value. With the initial value at 28%, Stata iterates to 18.57168. I conclude that 18.6% is the suitable breakpoint for the piecewise linear regression for the 12-month peak price. The breakpoints for the other peak prices are
determined in an equal way, with a non-linear regression in Stata. Again, I determine the initial values for the breakpoints by visual inspection of the gaussian kernel regressions.
Table 7 displays the breakpoints. I perform a likelikhood ratio test to analyse whether the model with a breakpoint is superior to the model without a breakpoint. For all peak prices, the test shows that
0 20 30 50 60 70 80 10 40 O ff e r p re mi u m 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
12 - month previous high
0 20 30 50 60 70 80 10 40 O ff e r p re mi u m 0 10 20 30 40 50 60 70 80 90 100 110 120 130 140
23
the model with a breakpoint is superior to the model without a breakpoint (p=0.000). Column 3 in table 7 provides the tests statistics which follow a chi-squared distribution.
Table 7 – Breakpoints piecewise linear regressions
The breakpoints determined in Stata with the non-linear command. Column 3 and 6 provide the test statistic for the test that the model with a breakpoint is equal to the model without a breakpoint. For all breakpoints, this hypothesis is rejected. The test statistic follows a chi-squared distribution.
Peak price period Breakpoint Test statistic Peak price period Breakpoint Test statistic
2 - month 28.4 17.91 14 - month 18.8 125.56 3 - month 13.8 35.66 15 - month 19.1 130.55 4 - month 18.2 57.96 16 - month 19.2 135.59 5 - month 19.3 68.26 17 - month 19.8 130.51 6 - month 18.2 87.39 18 - month 19.8 128.85 7 - month 17.8 104.56 19 - month 19.2 122.92 8 - month 18.0 109.01 20 - month 19.5 117.12 9 - month 18.2 123.22 21 - month 27.2 112.93 10 - month 18.0 111.87 22 - month 28.3 113.63 11 - month 18.1 111.78 23 - month 24.3 113.49 12 - month 18.6 115.96 24 - month 26.5 111.11 13 - month 18.2 116.48
Baker et al. (2012) and Niinivaara (2010) use 2 breakpoints. The first breakpoint of Baker et al. (2012) is at 25%, which is slightly higher than most of my breakpoints for the different past peak periods. They use an additional breakpoint at 75%. However, they do not find a significant effect after 25%. Niinivaara (2010) uses breakpoints at the 35% and 70% level. He does not find a significant effect after 35%. It should be noted that both Baker et al. (2012) and Niinivaara (2010) base the decision to set the breakpoints purely on visual inspection. They do not perform statistical tests to confirm their expectations.
I run the non-linear test in Stata again to test for an additional breakpoint. For 9 past peak prices, I find a second breakpoint that is significant at the 5% level. I run a likelihood-ratio test to compare the model with and without the additional breakpoint. The likelihood-ratio test is unable to reject the hypothesis that the model with 2 breakpoints is equal to the model with 1 breakpoint.
Based on the results of the likelihood-ratio test and the findings of Baker et al. (2012) and Niinivaara (2010), I continue the research with one breakpoint per past peak period.
24
conclude that the inclusion of the return on assets does not enhance the model. The AIC and BIC ratios, which are displayed in appendix A.2, are higher with the ROA variables included.
Without the restriction that the ROA variables are available, the sample size increases from 2,447 to 4,441.
Table 8 shows the results for the piecewise linear regression with the breakpoint determined before. The table displays the results for the 2, 6 and 12-month peak price period. Appendix A.3 displays the results including controls for all peak price periods. β1 accounts for the marginal effect of past peak prices up to the breakpoint. β2 accounts for the marginal effect of past peak prices beyond the breakpoint. With the inclusion of the controls, β2 is no longer significantly different from zero. This shows that a past peak price is no longer relevant when the share price has fallen substantially since the peak price. There are reasons such as a financial distress why a lower offer price is reasonable under such circumstances.
The 9-month past peak price has the highest β1(0.536) which means that this past peak price is strongest related to the offer price. The 8-month past peak price has a β1 of (0.511) which is not statistically different at the 10% level from the beta of the 9-month past peak price. β1 for the 8 and 9 month past peak price is statistically higher than many other past peak prices such as the 12-month past peak price. The hypothesis that β1 is equal for all past peak periods is rejected at the 1% level. The finding that the 8 and 9 month past peak prices have the greatest impact on the offer price suggests that bidders focus on the 8 and 9-month past peak price in the determination of the offer price.
The dummy variable cash is positive. This is in line with the expectations based on the fact that shareholders cannot defer the payment of taxes and do not benefit from potential future profits when they are paid in cash (de La Bruslerie, 2013).
The insignificant coefficient for the book-to-market of the target is in line with the literature. The literature is inconclusive about the sign. Comment (1995), Niinivaara (2010) and Schwert (2000) also find an insignificant sign for this variable.
I did expect a positive sign for the book-to-market of the bidder since bidders with higher growth opportunities pay higher offer premiums (Officer, 2003). Equal to Baker et al. (2012) and Niinivaara (2010), I do not find a significant coefficient for the book-to-market of the bidder.
The negative coefficient for the MCAP of the target confirms the expectations. This negative
25
potential economic benefits. Additionally, it is harder to achieve the assumed economic benefits due to the complexity of large targets.
Table 8 Piecewise linear regression
Regression of the offer premium on the 2, 6 and 12 month past peak prices. I run piecewise linear regressions with robust standard errors. The offer premium is the offer price from Zephyr expressed as a log percentage difference from the share price 30-days prior to the announcement date from DataStream. The x - month high is the highest share price of company i between day t-x and 31 days prior to the announcement bid, expressed as a log percentage difference from the share price 30-days prior to the announcement date. β1 accounts for the marginal effect of past peak prices up to the breakpoint and β2 for the marginal effect beyond the breakpoint. Cash (shares) is a dummy variable that is coded 1 when the offer includes a payment in cash (shares).
(1) (2) (3) (4) (5) (6) VARIABLES 2-month high 6-month high 12-month high 2-month high with controls 6-month high with controls 12-month high with controls β1 0.569*** 0.513*** 0.504*** 0.407*** 0.483*** 0.468*** (0.0729) (0.0657) (0.0615) (0.0717) (0.0639) (0.0601) β2 -1.527 0.114*** 0.0674*** -3.129 0.0238 0.0154 (2.722) (0.0386) (0.0188) (2.628) (0.0391) (0.0201) Inverse P 0.158* 0.112 0.0846 0.00738 0.00940 0.00556 (0.0939) (0.0945) (0.0943) (0.101) (0.101) (0.100) Cash 7.372*** 7.770*** 7.695*** (1.000) (1.008) (1.007) Shares -0.675 -0.790 -0.732 (0.907) (0.904) (0.904) Target BM 0.00846 0.00849 0.00749 (0.00741) (0.00740) (0.00740) Bidder BM -0.0134 -0.0122 -0.0127 (0.0107) (0.0108) (0.0108) Target MCAP -4.774*** -4.305*** -4.314*** (0.802) (0.808) (0.809) Bidder MCAP 5.453*** 5.420*** 5.455*** (0.682) (0.683) (0.684) VOL 2.278*** 2.204*** 2.154*** (0.276) (0.278) (0.289) Constant 20.71*** 18.43*** 17.10*** -0.983 -7.091 -7.975 (0.527) (0.637) (0.695) (6.482) (6.597) (6.628) Observations 4,441 4,441 4,441 4,441 4,441 4,441 R-squared 0.029 0.037 0.039 0.109 0.115 0.114 F 29.23 46.24 52.64 39.03 44.00 43.66
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
26
The positive coefficient for the volatility is in line with the findings of Baker et al. (2012) and our expectations. Offer premiums and the share price volatilities are both higher when investors are bearish. This link results in a positive relation.
Impact of peak prices on offer prices inside and outside the United States.
To test hypothesis 5, I run a sub-sample test which consists of offers for which the target is listed in the United States.
I run a Gaussian kernel regression to estimate the initial breakpoints for the non-linear regressions in stata. The breakpoints for this sub-sample are displayed in Appendix A.4. The only model that is statistically significant enhanced (at the 5% level) with a second breakpoint is the model for the 15-month past peak price. This model has a breakpoint at 8% and 40%. Since all other models are with 1 breakpoint, I continue the research with only one breakpoint. Table 9 displays the results of the regression of the piecewise regression including control variables for the 2, 6 and 12-month past peak price. Column 1 to 3 display the results for targets that are listed within the United States, Column 4 to 6 display the results for targets that are not listed in the United States.
The results for targets that are listed in the US are discussed first. β1 is positive and significant for all past peak prices. β2 is negative and significant for the 12 to 24-month past peak price. We expected a lower coefficient for β2 than for β1. The negative coefficient for the 12 to 24-month high insists that a past peak price is no longer relevant when the share price has fallen substantially since the peak price. There are reasons such as a financial distress why a lower offer price is reasonable under such circumstances.
27
Table 9 Piecewise linear regression for US and non-US targets
Piecewise linear regression for sub-sample tests. Columns 1 to 3 show the results of the regression which includes offers for which the target company is listed in the United States. Columns 4 to 6 show the results of the regression which includes offers for which the target company is not listed in the United States. β1 accounts for the marginal effect of premiums up to the breakpoint and β2 for premiums beyond the breakpoint.
(1) (2) (3) (4) (5) (6) VARIABLES 2-month high Only US 6-month high Only US 12-month high Only US 2-month high outside US 6-month high outside US 12-month high outside US β1 1.086*** 0.738*** 0.635*** 0.376*** 0.402*** 0.329*** (0.187) (0.104) (0.0750) (0.127) (0.0918) (0.0877) β2 -0.114 -0.0353 -0.0578* 0.267 0.0642 0.0541** (0.163) (0.0570) (0.0329) (0.264) (0.0503) (0.0255) Inverse P -0.241 -0.223 -0.206 0.130 0.128 0.123 (0.368) (0.373) (0.352) (0.104) (0.103) (0.103) Cash 5.815*** 5.930*** 5.825*** 7.109*** 7.558*** 7.525*** (1.572) (1.578) (1.578) (1.322) (1.334) (1.330) Shares -0.243 -0.298 -0.000871 -0.879 -0.993 -0.952 (1.197) (1.199) (1.188) (1.361) (1.353) (1.357) Target BM 0.0192 0.0185 0.0180 0.00228 0.00260 0.00126 (0.0119) (0.0119) (0.0118) (0.00926) (0.00925) (0.00925) Bidder BM -0.0159 -0.0125 -0.00908 -0.00752 -0.00735 -0.00944 (0.0179) (0.0180) (0.0182) (0.0132) (0.0132) (0.0132) Target MCAP -4.174*** -3.873*** -4.156*** -5.572*** -5.081*** -4.970*** (1.189) (1.187) (1.181) (1.100) (1.113) (1.118) Bidder MCAP 3.925*** 3.878*** 3.961*** 5.857*** 5.871*** 5.910*** (0.980) (0.983) (0.986) (0.949) (0.947) (0.945) VOL 2.841*** 2.788*** 2.895*** 1.618*** 1.547*** 1.415*** (0.421) (0.428) (0.443) (0.361) (0.364) (0.378) Constant 7.016 3.353 3.369 3.039 -3.300 -4.286 (10.22) (10.20) (10.11) (8.943) (9.150) (9.254) Observations 1,888 1,888 1,888 2,553 2,553 2,553 R-squared 0.156 0.160 0.166 0.079 0.083 0.082 F 23.67 26.04 28.68 17.42 19.13 18.64
Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
28
Columns 4 to 6 display the regression results for targets that are not listed in the United States. β1 is most of the time significantly higher (at 1% level) in the US. This finding shows that bidders are more prone to use past peak prices for targets that are listed in the US than for targets that are listed in other countries. β1 is not significantly different in and outside the US for the 15-month and 17 to 24 month peak prices. I find limited proof for the use of past peak prices for targets that are listed outside the US.
I conclude that hypothesis 5a can be accepted. The past peak prices have a greater impact on the offer premiums for targets that are listed in the United States.
The signs of the control variables are equal inside and outside the United States. See page 25 for a discussion of the control variables.
Deal success
29
Table 10 Difference success rate above/below past peak price
Panel A displays descriptive statistics for offers that are recorded as completed in the Zephyr database. Panel B displays the difference in success rate between premiums that do exceed and do not exceed the past peak price. “Below” corresponds to the deal success rate of offer prices which are below the past peak price. “Above” corresponds to offers that exceed the past peak price. Delta is the difference between columns Below and Above i.e. the difference in percentage of completed offers between offers that exceed the past peak price and do not exceed the past peak price.
Panel A
Completed Not completed Full sample
N 8,830 2,601 11,431
% 77% 23% 100%
Median offer premium 21.70 18.50 20.91
Average offer premium 24.95 22.11 24.30
Panel B
Peak price period Below Above Delta Peak price period Below Above Delta
2 - month 73.7% 78.2% 4.5% 14 - month 76.1% 78.4% 2.4% 3 - month 73.8% 78.4% 4.5% 15 - month 76.0% 78.5% 2.5% 4 - month 74.3% 78.4% 4.1% 16 - month 76.1% 78.5% 2.4% 5 - month 75.0% 78.3% 3.3% 17 - month 76.2% 78.4% 2.2% 6 - month 75.3% 78.3% 3.0% 18 - month 76.3% 78.3% 2.0% 7 - month 75.4% 78.3% 3.0% 19 - month 76.4% 78.3% 1.9% 8 - month 75.7% 78.3% 2.6% 20 - month 76.4% 78.3% 1.8% 9 - month 75.9% 78.2% 2.3% 21 - month 76.5% 78.2% 1.7% 10 - month 75.8% 78.4% 2.6% 22 - month 76.3% 78.5% 2.2% 11 - month 75.9% 78.4% 2.5% 23 - month 76.3% 78.5% 2.2% 12 - month 75.9% 78.4% 2.5% 24 - month 76.3% 78.6% 2.2% 13 - month 76.0% 78.4% 2.4%
Following Baker et al. (2012), I run a probit regression on deal success. I split the sample in two samples; one sample includes offers for which the target company is listed outside the United States, the other includes offers for which the target company is listed in the United States. I split the sample since table 9 suggests that past peak prices play a different role inside and outside the US.
Table 11 shows the results of the regression of deal success on the offer premium for targets listed outside the US. Columns 1 to 3 show the results of the regression for the 2, 6 and 12-month past peak price including control variables. Columns 4 to 6 include the controls and the polynomial to the power 2, 3 and 4 of the offer premium. The polynomials will load on everything that is related to a
continuous increase in deal success. When the offer premium dummy π is still positive and
30
For none of the 23-past peak prices, the offer premium dummy π is significant. This means that I do not find proof for a discontinuous jump in deal success when the offer price exceeds the past peak price for companies that are listed outside the US. The results of all peak price periods are displayed in Appendix A.4
Table 11 Jump in deal success outside the United States
Results of the probit regression run as Ω𝒊= 𝜶 + 𝜷(𝝅𝒊), + 𝝐 𝒊 where Ω is a dummy variable coded 1 for completed offers
and 0 for not completed offers. π is a dummy variable that is coded 1 when the offer price exceeds the past peak price and 0 otherwise. Columns 1 to 3 control for the type of deal payment, target market cap (MCAP) and bidder market cap. Columns 4 to 6 include the polynomials to the power 2, 3 and 4 of the offer premium. A discontinuous jump in deal success when the
past peak price is exceeded will be visible by a significant coefficient for 𝝅𝒊. Only offers for which the target is not listed in
the United States are included.
(1) (2) (3) 5 (5) (6) VARIABLES 2-month high 6-month high 12-month high 2-month high 6-month high 12-month high Offer premium -0.000438 -0.000734 -0.000554 -0.00509 0.00680** 0.00626** (0.00114) (0.00109) (0.00102) (0.00388) (0.00325) (0.00308) Offer premium² 0.000258*** 0.000261*** 0.000259***
(7.94e-05) (7.98e-05) (7.99e-05)
Offer premium³ 1.07e-05*** 1.15e-05*** 1.12e-05***
(3.36e-06) (3.27e-06) (3.25e-06)
Offer premium4 -8.48e-08*** -9.09e-08*** -8.89e-08***
(2.81e-08) (2.72e-08) (2.71e-08)
π 0.0228 -0.00845 0.0126 0.0366 -0.0324 -0.00975 (0.0734) (0.0638) (0.0578) (0.0952) (0.0688) (0.0602) Inverse P 0.00672*** 0.00666*** 0.00672*** -0.00657** -0.00649** -0.00654** (0.00255) (0.00255) (0.00255) (0.00258) (0.00258) (0.00258) Cash -0.305*** -0.308*** -0.305*** -0.308*** -0.314*** -0.311*** (0.0621) (0.0624) (0.0621) (0.0624) (0.0627) (0.0626) Shares 0.185*** 0.186*** 0.185*** 0.177*** 0.180*** 0.179*** (0.0622) (0.0623) (0.0623) (0.0623) (0.0624) (0.0624) Target MCAP -0.420*** -0.422*** -0.420*** -0.431*** -0.434*** -0.433*** (0.0464) (0.0465) (0.0467) (0.0467) (0.0469) (0.0471) Bidder MCAP 0.284*** 0.284*** 0.284*** 0.282*** 0.282*** 0.282*** (0.0411) (0.0411) (0.0411) (0.0412) (0.0412) (0.0412) Constant 1.773*** 1.806*** 1.773*** 1.990*** 2.062*** 2.034*** (0.290) (0.295) (0.299) (0.297) (0.301) (0.305) Observations 3,417 3,417 3,417 3,417 3,417 3,417 Pseudo R-squared 0.0396 0.0396 0.0396 0.0440 0.0440 0.0440 chi2 139.7 139.7 139.8 154.9 154.9 154.8
31
I find that deals paid in cash are less likely to be accepted. This is in contrary to my expectations based on the fact that stock swap deals require approval from the shareholders of the bidders as well. The negative coefficient for cash offers is in line with the findings of Baker et al. (2012). This finding suggests that bidders are more likely to accept an offer when they are paid in shares since they can benefit from potential future profits.
The coefficient for target market cap is negative and significant at the 1% level. The bidder market cap is positive and significant at the 1% level. These findings are in line with my expectations.
Deal success in the US
Table 12 shows the results for the probit regression of the 2, 6 and 12-month past peak price for targets that are listed in the US.
Table 12 Jump in deal success for US Targets
Results of the probit regression run as Ω𝒊= 𝜶 + 𝜷(𝝅𝒊), + 𝝐 𝒊 where Ω is a dummy variable coded 1 for completed offers
and 0 for not completed offers. π is a dummy variable that is coded 1 when the offer price exceeds the past peak price and 0 otherwise. Only significant control variables are included. A discontinuous jump in deal success when the past peak price is
exceeded will be visible by a significant coefficient for 𝝅𝒊. Only offers for which the target is listed in the United States are
included.
(1) (2) (3)
VARIABLES 2-month high 6-month high 12-month high Offer premium 0.0185*** 0.0171*** 0.0172***
(0.00333) (0.00258) (0.00244) Offer premium2 -0.000241*** -0.000233*** -0.000229***
(3.81e-05) (3.25e-05) (3.27e-05)
π 0.0569 0.158* 0.184** (0.129) (0.0853) (0.0727) Inverse P 0.00817 0.00868 0.00981 (0.00713) (0.00732) (0.00736) Shares 0.139** 0.151** 0.151** (0.0686) (0.0686) (0.0690) Target MCAP -0.394*** -0.405*** -0.409*** (0.0611) (0.0613) (0.0615) Bidder MCAP 0.359*** 0.360*** 0.360*** (0.0527) (0.0527) (0.0525) Constant 0.725* 0.762* 0.803** (0.398) (0.396) (0.398) Observations 2,451 2,451 2,451 Pseudo R-squared 0.0632 0.0648 0.0662 chi2 128.1 131.6 136.0
32
Appendix A.5a displays the regression results for all past peak prices. The regression excludes variables Offer premium³, Offer premium4 and Cash for being insignificant. Appendix A.5b displays the regressions which include the variables Offer premium3, Offer premium4 and Cash.
The discontinuous jump in deal success when the past peak price is exceeded (displayed in π) is positive and significant for all past peak prices except the 2, 4 and 5-month past peak price. The jump I find is between 15% and 25.8%.
Baker et al. (2012) execute this test for the 12-month past peak price. They find a discontinuous jump in deal success (π) of 6% when the offer exceeds the 12- month past peak price. I find a significantly higher (at 10% level) π of 18.4% when the 12-month past peak price is exceeded.
This finding shows that there is a discontinuous jump in deal success when the offer price exceeds the past peak price.
The Wald test that tests the hypothesis that the π’s for the different past peak prices are all equal provides a p-value of 0.45. Therefore, I do not find proof that there is a single past peak price that has a special role in the reception of the offer price by the shareholders of the target.
Subsample and robustness checks
The results of Niinivaara (2010) suggest that shareholders of targets that are listed in Europe use past peak prices as reference points in the reception of the offer. In my earlier tests I do not find proof for the use of past peak prices for targets that are listed outside the United States.
I run a sub-sample test in which I only include targets that are listed in Europe. In this sample, I do not find proof for the use of past peak prices among shareholders of targets that are listed in Europe. My results conflict with the results of Niinivaara (2010) and suggest that the (lack of) shareholder concentration is important. The results of my tests are displayed in appendix A.6.
33
5. Conclusion and limitations
In this thesis, I analyse the impact of past peak prices on the height of offer prices and deal success. My sample consists of 11,431 offers that are announced between 2000 and 2015.
I do find a positive and significant relation between the past peak price and the offer price. This confirms the findings of findings of Baker et al. (2012) and Niinivaara (2010) and is in line with Hypothesis 1.
My results are in line with hypothesis 2; I find that the relation between the offer premium and peak price is non-linear. The impact of the peak price on the offer premium is stronger when the peak price is relatively low.
In line with hypothesis 5, I find that the impact of the past peak prices on offer premiums is stronger for companies that are listed in the United States (US). This finding supports the view that bidders anticipate on the use of past peak prices as reference points by shareholders of firms that are listed in the US. My results suggest that the bidders specifically focus on the 2-month peak price. I have to reject hypothesis 4a which states that the impact of the 52-week peak price is greater than any other past peak price.
Hypothesis 3 tests whether the shareholders of the targets use past peak prices as reference prices. The results I find are blended. I find proof for a jump in deal success when the offer price exceeds the peak price. However, this is only the case for targets that are listed in the US. The shareholders of those targets do not focus on a specific past peak period. I do not find a jump in deal success for targets that are listed outside the US.
My results regarding deal success conflict the results of Niinivaara (2010). He finds a discontinuous jump in deal success in his sample of European offers between 1997 and 2008. I do not find proof for a discontinuous jump in deal success in the sub-sample which only includes targets that are listed in Europe.
I find proof for the use of past peak prices as reference points by shareholders, but only for targets that are listed in the US. I do not find proof that shareholders focus on a single past peak period. Therefore, I cannot accept hypothesis H4b which states that the 52-week peak price has the greatest impact on deal success.
34
35
References
Alexandridis, G., Fuller, K., Terhaar, L., Travlos, N., 2013. Deal size, acquisition premia and shareholder gains. Journal of Corporate finance 20, 1-13.
Altman, N., 1992. An introduction to kernel and nearest-neighbor nonparametric regression. The American Statistician 46, 175-185.
Baker, M., Pan, X., Wurgler, J., 2009. A reference point theory of mergers and acquisitions. Unpublished Working Paper. Harvard University and New York University.
Baker, M., Pan, X., Wurgler, J., 2012. The effect of reference point prices on mergers and acquisitions. Journal of Financial Economics 106, 49-71.
Branch, B., Wang, J., Yang, T., 2008. A note on takeover success prediction. International review of financial analysis 17, 1186-1193.
De La Bruslerie, H., 2013. Crossing takeover premiums and mix of payment: an empirical test of contractual setting in M&A transactions. Journal Of Banking & Finance 37, 2106-2123.
Chira, I., Madura J., 2015. Reference point theory and the pursuit of deals. The Financial Review 50 50, 275-300
Comment, R., Schwert., W., 1995. Poison or placebo? Evidence on the deterrence and wealth effects of modern antitakeover measures. Journal of financial economics 39, 3-43.
DSE Unibo. 2016. Databases. (Accessed March 8, 2016)
Retrieved from http://www.dse.unibo.it/en/libraries/databases/databases/#zephyr
Franks, J., Mayer, C., Rossi, S., 2009. Ownership: evolution and regulation. The Review of Financial Studies 22, 4009-4056.
Georgen, M., Renneboog, L., 2004. Shareholder wealth effects of European domestic and cross-border takeover bids. European financial management 10, 9-45.
Ghosh, D., Vogt, A., 2012. Outliers: An Evaluation of Methodologies. Section on Survey Research Methods - JSM 2012, 3455-3460
36
Kahneman, D., Tversky, A., 1979. Prospect theory: an analysis of decision under risk. Econometrica 47, 263-292.
KPMG,. 2015. Spotlight on capital allocation: M&A may be the best use of capital. Retrieved from https://advisory.kpmg.us/deal-advisory/ma-spotlight/ma-spotlight-december-2015.html
Niinivaara, T., 2010. Role of psychological reference points in mergers and acquisitions: 52-week high as a reference price in European takeover activity. Unpublished working paper. Aalto University, Finland.
Offenberg, D., Pirinsky, C., 2014. How do acquirers choose between mergers and tender offers? Unpublished working paper. Loyola Marymount University., George Washington University. Officer, M., 2003. Termination fees in merger and acquisitions. Journal of financial economics 69, 431- 67.
Petitt, B., Ferris, K., 2013. Valuation for mergers and acquisitions. 2nd edition. Pearson education, New Jersey.
Roll, R., 1986. The hubris hypothesis of corporate takeovers. The journal of Business 59, 179-216. Ryan, S., Porth, L., 2007. A tutorial on the piecewise regression approach applied to bedload
transport data. United States Department of Agriculture & Forest Service Rocky Mountain Research Station, Fort Collins, CO, US.
Schwert, G., 2000. Hostility in takeovers: In the eyes of the beholder? The journal of finance 45, 2599-2640.
Shefrin, H., Statman, M., 1985. The disposition to sell winners too early and ride losers too long: theory and evidence. Journal of Finance 40, 777-790.
Simonyan, K., 2014. What determines takeover premia: an empirical analysis. Journal of economics and business 75, 93-125.
StataCorp, 2015. Stata 14 Base Reference Manual. College Station, TX: Stata Press.
37
UCLA: Statistical Consulting Group., 2016. Stata Data Analysis Examples: Probit Regression (accessed May 5, 2016)
Retrieved from http://www.ats.ucla.edu/stat/stata/dae/probit.htm
UCLA: Statistical Consulting Group., 2016. Stata FAQ: How can I find where to split a piecewise regression? (accessed May 22, 2016)
Retrieved from http://www.ats.ucla.edu/stat/stata/faq/nl_optimal_knots.htm
UCLA: Statistical Consulting Group., 2016. Regression with Stata Chapter 4 - Beyond OLS (accessed May 23, 2016)
