The performance and impact of stock picks mentioned on Mad Money

Transcription

1 Applied Financial Economics, 2010, 20, The performance and impact of stock picks mentioned on Mad Money Bryan Lim a, * and Joao Rosario b a Department of Finance, University of Melbourne, Melbourne, VIC 3010, Australia b Department of Economics, University of California, Santa Barbara, CA, USA We analyse both the market reaction and the long-term returns of stock picks mentioned on the Consumer News and Business Channel (CNBC) programme Mad Money, hosted by former hedge fund manager Jim Cramer. We find that Cramer s stock-picking style is consistent with a positive-feedback trading strategy, favouring stocks which have outperformed over an interval prior to the pick date. Subsequent to a pick, Cramer s immediate effect on a stock appears inversely proportional to the corresponding firm s market capitalization. The returns over a 6-month horizon provide some evidence in favour of Cramer s stock-picking ability. In particular, his recommendations on small-cap stocks accurately predict the long-run trends. I. Introduction Nearly every business day at 6 pm Eastern Standard Time, former hedge fund manager Jim Cramer hosts an hour-long show on the cable financial network Consumer News and Business Channel (CNBC). Typically reaired once again at 11 pm, the programme is billed as a primer on personal investment, with Cramer offering his recommendations and criticisms of companies in an effort to, in his words, make you money. While historically there has never been a scarcity of financial prognosticators willing to offer their wisdom to anyone who will listen, few if any can claim the breadth of the domain Cramer commands, with up to viewers daily. That his emergence coincided with the ever-increasing ubiquity of personal trading via the internet raises the possibility that where other Wall Street soothsayers have been rebuffed by market forces, Cramer and his viewership are market forces unto themselves. Were his audience both adequately numerous and dedicated, Cramer s picks at least in the immediate horizon would become self-fulfilling prophesies: a glowing recommendation on the show followed by a swell of buy orders the next trading day would provide instant vindication. To form a complete analysis of his recommendations, then, one needs to separate his influence from his forecasting ability. To this end, we evaluate the market response immediately following a broadcast of Mad Money as well as Cramer s ability to forecast winners and losers over a 6-month horizon. We analyse recommendations on the Mad Money programme along three lines of inquiry. First, we look for the evidence of positive-feedback trading in Cramer s stock-picking style by examining the trends of stocks prior to recommendations. Second, we estimate the immediate impact of recommendations *Corresponding author. blim@unimelb.edu.au Applied Financial Economics ISSN print/issn online ß 2010 Taylor & Francis DOI: /

2 1114 B. Lim and J. Rosario by the market reaction subsequent to a broadcast. Finally, we test his forecasting ability by calculating the long-term (420 trading days) returns associated with recommendations. We find that Cramer s stockpicking style is consistent with a positive-feedback trading strategy, favouring stocks which have outperformed over an interval prior to the pick date. Subsequent to a pick, Cramer s immediate effect on a stock appears inversely proportional to the corresponding firm s market capitalization. This result is not unexpected; if the same number of viewers responds to each pick, the disturbance on a smallcap stock will be relatively greater than that on a large-cap stock. The returns over a 6-month horizon provide some evidence in favour of Cramer s stockpicking ability. In particular, his recommendations on small-cap stocks accurately predict the long-run trends. Cramer professes no insider knowledge, frequently exhorting his audience to research firms guidance (earnings estimates) before investing. While the spikes and valleys in the short-term returns of his picks are likely due to a substantial portion of his audience following his advice, over a sufficiently long horizon we would expect the market to adjust given that these price changes contain no new information. As such, we interpret the long-term excess returns associated with Cramer s small-cap stock picks as potential evidence of his skill, as opposed to influence. Rudimentary analyses in the nonacademic press typically find little favourable evidence of Cramer s ability. We postulate here that such results are driven by the fact that nearly 75% of Cramers picks represent either large-cap stocks or responses to callers questions or both. In general, size-adjusted excess returns for large-cap stocks are likely to be relatively small, in which case it may not be possible for Cramer to generate significant excess returns within this category. By the same token, the market response to caller picks is small versus the response to Cramer s in-depth recommendations. More than likely, the amount of information (that is perceived to be) conveyed to viewers during the caller segments is relatively low. Taken together, the huge numbers of large-cap and caller picks are likely weighting the calculated excess returns towards zero. This article is similar in spirit to two contemporaneous papers. Engleberg, Sasseville and Williams (ESW, 2007) examine first-time picks on the Mad Money programme. Concentrating on a 6-month window with 391 picks, they find significant next-day returns for first-time positive recommendations by Cramer, returns which increase inversely with market capitalization. By comparison, we document smaller excess returns subsequent to recommendations, which may be attributable to either the differing sample selection or the differing measurement of excess returns. Similar to our article, Keasler and McNeil (2010) document returns associated with Cramer s picks, with a sample period approximately 6 months shorter than ours. While both their and our calculations for short-term excess returns are comparable, we find limited evidence of positive excess longer-term returns while Keasler and McNeil do not. This article is further differentiated by our stressing trends in Cramer s stock picking style as well as in the overnight component of the 1 day returns associated with his picks. More generally, this article represents an intersection of the event study and analyst recommendation literatures. The analysis most closely follows Womack s (1996) examination of both the market reaction and the long-term returns associated with analyst recommendations. The mass-media component of this article is in the spirit of research documenting the market reaction to stocks mentioned in financial news outlets. Barber and Loeffler (1993), Greene and Scott (1999) and Liang (1999) examine the market response to the Wall Street Journal s Dartboard column. Lloyd-Davies and Canes (1978) and Liu et al. (1990) similarly examine the market response to the Wall Street Journal s Heard on the Street column. Busse and Green (2002) document intraday trading response to CNBC s Morning Call and Midday call segments, while Barber and Odean (2006) examine the effects of high levels of news coverage on a given stock s price. II. Data Our sample consists of episodes broadcast between 28 June 2005 and 22 December The data was taken from official recaps posted at thestreet.com, a financial website founded by Cramer. Our transcribed results were spot-checked against independent recap websites, 1 though none had picks for shows earlier than November We restrict tickers to those listed on the New York Stock Exchange (NYSE), National Association of Securities Dealers Automated Quotations (NASDAQ) or American Express (AMEX) exchanges. Picks are identified as either Buy or Sell. While this simple binary categorization effectively eliminates distinguishing between weak and strong 1 and

3 Mad Money 1115 recommendations (positive or negative), it also removes the subjectivity that would be involved in such a transcription process. In this sense, a Buy may not represent the actual word buy being spoken by Cramer but simply an endorsement, however measured. We further separate picks by the party who initiated the ticker in question: caller or noncaller. During such segments as Lightning Round, Sudden Death and Are You Diversified?, viewers call into the programme and ask Cramer for his opinion on particular stocks. Caller picks are typically discussed only cursorily, and it seems unlikely that on average a 5-second recommendation during the Lightning Round will convey the same information as a 5-minute recommendation during the opening segment. In terms of volume, the vast majority of the tickers mentioned on the programme represent questions from callers. Cramer ostensibly has no previous knowledge that these tickers will be inquired about (though they are often follow-ups to his previous picks), and as such, his responses may not fairly represent his stock-picking ability. The majority of a broadcast s running time consists of Cramer discussing what we call noncaller picks. These picks represent firms that Cramer premeditatedly chooses to mention on the programme. Unlike caller picks, he has complete discretion over the firms in this category. Regular viewers of Mad Money will observe the absence of other potential categorizations, like Pick of the Week or Sell Block. While such categories appear in our database, given the evolving nature of the programme we have chosen only the most general categories in which to sort. For each trading day, we rank each stock of the combined NYSE, NASDAQ and AMEX databases to determine its market capitalization decile on that date. We group stocks into three bins: small cap for deciles 1 5, mid cap for deciles 6 8; and large cap for deciles 9 and 10. We stress that these classifications are relative and therefore do not correspond to the standard definitions in the practitioner s lexicon. Our primary sample consists of picks exclusively from first-run episodes. Table 1 summarizes the data. The picks represent 2074 distinct firms from 560 unique Standard Industrial Classification (SIC) codes. (Remarkably, over the sample period, the entire Center for Research in Security Prices (CRSP) database lists 8012 distinct firms from 922 unique SIC codes.) Examining Table 1, we observe several trends:. Among noncaller picks, Buys outnumber Sells by more than 5 to 1. This ratio is roughly fixed across each market capitalization group. By and large, the programme is a forum for Cramer to highlight companies he prefers as opposed to those he dislikes. There may be any number of reasons to explain this tendency, with a likely culprit being his audience s inability or unwillingness to sell negatively recommended stocks. In Section Overnight returns, we document only modest market reaction to a Sell pick. It may be the case that Cramer simply tailors his show to match the investing habits of his audience.. Among caller picks, the probability of Cramer recommending a Buy increases with market capitalization. For caller tickers from deciles 1 to 5, the Buy-to-Sell ratio is less than 1 to 2, while for tickers in deciles 9 and 10, the ratio switches to about 2.3 to 1.. Approximately two-thirds of all picks represent firms from the top 20% market capitalization. This ratio holds for both noncaller and caller picks. Table 1. Summary statistics Market cap decile All and 10 Distinct tickers Distinct SIC codes Noncaller picks Buys Sells Caller picks Buys Sells Full sample Average market cap $ $227 $880 $ Note: Market capitalization in millions.

4 1116 B. Lim and J. Rosario III. Pre- and Post-pick Returns We calculate the raw, size-adjusted and industryadjusted excess returns for the stocks mentioned on the programme. As Mad Money is broadcast after markets close, we must be precise in our holdingperiod definitions. Given a pick date T and a holding period before or after T, we calculate the following time-horizon (net) returns:. Period before pick date ¼ (Closing price on T /Closing price on T ) 1. Period after pick date: (Closing price on T þ / Closing price on T ) 1 All returns are calculated cum dividend, with the exception of the overnight returns discussed later in this section. There is no distinction between calculating returns for Buys and for Sells; they are computed identically. Specifically, returns are calculated per stock, so a positive return on a Buy or a Sell simply means that the stock itself has outperformed its benchmark over the specified horizon. In terms of post-pick performance, positive (excess) returns for Buys and negative (excess) returns for Sells would ostensibly be desirable from a viewer s perspective. Raw returns We calculate the return to stock i over period as 1 þ R i, ¼ Y ð1 þ r i,t Þ t where r i,t is the return from holding stock i from the market close on date t 1 to the market close on date t. Size-adjusted excess returns We compare pick returns to the CRSP valueweighted index for each pick s market capitalization decile, ranked against the aggregated AMEX, NASDAQ and NYSE firms. The size-adjusted excess return over period is calculated as the return for firm i minus the return for the CRSP valueweighted index for the corresponding decile 1 þ ER size i, ¼ Y t ð1 þ r i,t Þ Y t Industry-adjusted excess returns ð1 þ r decileðiþ, t Þ We compare pick returns versus averages of other firms in the same industry. Following Womack (1996), we construct equally-weighted portfolios of firms with the same SIC code as the pick firm on the pick date (note that a firm s SIC code is not necessarily constant over time) for those with a minimum of five other firms with the same SIC. We calculate industry-adjusted excess returns over period as the size-adjusted excess return for firm i minus the average size-adjusted excess return for the J other firms with the same SIC code as firm i Caller picks ER SIC i, ¼ ER size i, 1 J X J j2sicðiþ j6¼i ER size j, As callers do not know ex ante whether Cramer will positively or negatively recommend their picks, we calculate the pre-pick returns for all caller picks, regardless of Cramer s eventual recommendation, in order to capture the information set as of the time of filming. The average size-adjusted excess returns over the 60 and 20 trading days immediately prior to the pick date are 6.78% (t ¼ 23.4) and 2.34% (t ¼ 13.7), respectively. That these returns are positive and significant suggests that Cramer s audience at least, those who call in is generally interested in his opinion on hot stocks, those that have outperformed over the previous month or quarter. Viewers presumably call in to gauge his opinion on whether the stock has peaked or not. Figure 1 graphs cumulative average size-adjusted excess returns for caller picks over an interval of 20 days prior and subsequent to the pick date, denoted by day 0. While both buys and sells exhibit increasing returns over the interval up to day 0, we observe that Cramer recommends buying stronger outperformers and selling weaker ones. More precisely, the average 20 trading-day pre-pick size-adjusted excess return on Caller picks Buys Sells Trading days relative to pick date Fig. 1. Cumulative size-adjusted excess returns for caller picks Note: The horizontal lines indicate the cumulative excess return as of the market close on the date of the pick.

5 Mad Money 1117 a caller ticker which Cramer recommends buying is 3.47%, while the equivalent for a sell is only 1.34%, as shown in Table 2. Looking at the post-pick period, we observe small, positive but significant excess returns for caller buys over the next possible trading day, gradually decaying towards zero over the following month. Similarly, for caller sells, we observe small, negative but significant excess returns over the next possible trading day, remaining relatively flat over the following month. The size-adjusted excess returns over a 20 trading-day horizon for buys and sells are not significantly different, a fact which may have more to do with both the breadth of the selections and the selection process of Cramer s audience than with Cramer himself. Conditional on market capitalization. We decompose the pre- and post-pick returns by market capitalization. Figure 2 graphs the results. Within each market capitalization group, callers inquire about stocks which are outperforming, with the level of outperformance roughly negatively related to the market capitalization. For small-cap stocks, the pre-pick cumulative excess returns of buys and sells are nearly identical, suggesting Cramer could not be applying any simple technical analysis rule to make his recommendations within this group. More interestingly, the post-pick returns for small-cap caller picks suggest that his recommendations for this group are correct: Cramer is on average correctly identifying whether or not the stock in question has reached its short-term peak. Small-cap caller sells exhibit a downward drift after the pick date, while the corresponding buys peak 7 trading days after the pick date before drifting downward. Moreover, of the small cap stocks he is asked by callers, his sells underperform while his buys outperform over the subsequent 20 trading days. The results for mid-cap caller picks are considerably less dramatic. A recommendation on the show elicits a small next-day excess return with the proper Table 2. Caller pick returns Size-adjusted Industry-adjusted Raw return Excess return Excess return Sells Buys Sells Buys Sells Buys All caller picks (n ¼ 2971) (n ¼ 4916) (n ¼ 2971) (n ¼ 4916) (n ¼ 2399) (n ¼ 3917) 20-Day period before pick (7.58) (24.62) (3.79) (17.37) (3.18) (12.86) 1-Day period after pick ( 5.50) (10.50) ( 7.13) (9.20) ( 6.79) (7.52) 20-Day period after pick (2.85) (6.35) ( 2.27) ( 0.77) ( 3.97) ( 1.60) (n ¼ 213) (n ¼ 93) (n ¼ 213) (n ¼ 93) (n ¼ 162) (n ¼ 74) 20-Day period before pick (5.46) (3.13) (4.59) (2.60) (4.22) (2.00) 1-Day period after pick ( 3.55) (4.99) ( 4.05) (4.62) ( 3.10) (3.75) 20-Day period after pick ( 1.35) (1.49) ( 2.33) (1.25) ( 3.05) (1.54) (n ¼ 1139) (n ¼ 1033) (n ¼ 1139) (n ¼ 1033) (n ¼ 892) (n ¼ 782) 20-Day period before pick (5.22) (12.34) (3.15) (9.61) (2.24) (8.07) 1-Day period after pick ( 4.28) (7.65) ( 5.30) (7.34) ( 4.73) (5.31) 20-Day period after pick (0.27) (1.12) ( 2.59) ( 1.32) ( 3.52) ( 2.14) (n ¼ 1620) (n ¼ 3790) (n ¼ 1620) (n ¼ 3790) (n ¼ 1346) (n ¼ 3061) 20-Day period before pick (2.75) (23.50) ( 1.80) (15.59) ( 1.03) (10.78) 1-Day period after pick ( 1.86) (6.29) ( 3.07) (4.84) ( 3.79) (4.51) 20-Day period after pick (5.47) (6.99) (0.86) ( 0.35) ( 0.48) ( 0.89) Note: t-statistics are given in parentheses.

6 1118 B. Lim and J. Rosario 0.18 Caller Sells 0.16 Deciles Deciles Caller Buys 0.16 Deciles Deciles Fig. 2. Cumulative size-adjusted excess returns for caller picks, conditional on market capitalization Note: The horizontal lines indicate the cumulative excess return as of the market close on the date of the pick. corresponding sign (positive for buys and negative for sells); subsequent to that first trading day, the postpick drift for sells is flat and for buys slopes downward. After 20 trading days, the post-pick excess returns for both mid-cap buys and sells are not significantly different from each other. Large cap stocks appear generally undisturbed by Cramer s recommendations to callers questions. While he favours large cap outperformers to underperformers, the post-pick drift for both large-cap caller buys and sells is essentially flat. Noncaller picks As observed in Fig. 3, among noncaller picks, the stocks that Cramer recommends selling have underperformed over the previous month while those he recommends buying have outperformed over the same horizon. Viewed alongside his pattern of recommending the highest outperformers among caller-requested stocks, these tendencies suggest that Cramer favours a positive feedback strategy: buying (relative) winners and selling (relative) losers (Table 3) Noncaller picks Buys Sells Trading days relative to pick date Fig. 3. Cumulative size-adjusted excess returns for noncaller picks Note: The horizontal lines indicate the cumulative excess return as of the market close on the date of the pick. Immediately after a recommendation, we observe significant next-day size-adjusted excess returns: 0.88% for sells and 1.56% for buys. Over the next month, however, these excess returns drift towards zero, though for buys, the excess returns remain positive. Conditional on market capitalization. Figure 4 graphs the cumulative size-adjusted excess returns for noncaller picks, decomposed by market capitalization. As evidenced by the pre-pick excess returns, Cramer tends to favour outperformers and to reject underperformers within each market capitalization group. Following a broadcast, the next-day excess returns for small-cap noncaller picks are relatively large in magnitude, particularly for buys. The excess returns for buys exhibit a slight upward drift after the first trading day, while for sells they exhibit a slight downward drift over the first 2 weeks (10 trading days) before sharply increasing thereafter. If purchasing the stock at the closing price of the next trading day after a broadcast, however, one earns significantly higher 1-month excess returns purchasing small-cap stocks Cramer recommends selling than those he recommends buying. Excess returns for mid-cap noncaller buys increase sharply the next trading day, then drift downward towards zero over the subsequent month. For sells, the next-day excess returns appear to be part of a general downward drift that begins 8 trading days prior to the pick date and ends 8 trading days afterward. Echoing the observed trends for large-cap caller picks, noncaller large-cap stocks are largely unaffected by a positive or negative recommendation by Cramer. Sells exhibit a temporary dip in excess

7 Mad Money 1119 Table 3. Noncaller pick returns Size-adjusted Industry-adjusted Raw return Excess return Excess return Sells Buys Sells Buys Sells Buys All noncaller picks (n ¼ 445) (n ¼ 2260) (n ¼ 445) (n ¼ 2260) (n ¼ 337) (n ¼ 1742) 20-Day period before pick (0.13) (17.91) ( 1.88) (13.76) ( 1.95) (9.46) 1-Day period after pick ( 3.91) (17.24) ( 4.29) (17.26) ( 4.87) (14.54) 20-Day period after pick (1.37) (8.85) ( 0.70) (4.17) ( 1.72) (3.99) (n ¼ 25) (n ¼ 110) (n ¼ 25) (n ¼ 110) (n ¼ 20) (n ¼ 88) 20-Day period before pick ( 1.16) (3.34) ( 1.19) (2.24) ( 1.19) (0.41) 1-Day period after pick ( 1.83) (7.22) ( 1.89) (7.16) ( 2.28) (6.23) 20-Day period after pick (0.41) (5.66) (0.17) (5.11) (0.77) (4.32) (n ¼ 139) (n ¼ 553) (n ¼ 139) (n ¼ 553) (n ¼ 95) (n ¼ 414) 20-Day period before pick ( 0.36) (9.44) ( 1.27) (7.92) ( 0.94) (6.22) 1-Day period after pick ( 2.26) (13.36) ( 3.09) (13.89) ( 3.27) (11.69) 20-Day period after pick ( 0.04) (2.94) ( 1.19) (1.29) ( 2.51) (1.16) (n ¼ 281) (n ¼ 1593) (n ¼ 281) (n ¼ 1593) (n ¼ 222) (n ¼ 1238) 20-Day period before pick (1.67) (15.99) ( 0.63) (11.76) ( 1.34) (7.84) 1-Day period after pick ( 2.86) (11.18) ( 2.84) (10.94) ( 2.94) (9.22) 20-Day period after pick (2.13) (7.49) ( 0.10) (2.40) ( 1.10) (2.44) Note: t-statistics are given in parentheses. returns, while buys exhibit a small but persistent increase. Overnight returns Given the daily nature of broadcasts, we separate the overnight component of the 1-day raw returns to examine the immediate reaction to picks on the show. Given a broadcast date t, we define the overnight return as r overnight ¼ Opening price on t þ 1 1 Closing price on t As shown in Table 4, the overnight change in price accounts for virtually all of the 1 day returns, regardless of the subsample. The difference in the next-day opening price from the previous-day close can be accounted for by either after-hours trading or an accumulation of orders in the specialists books before markets open. For two subsamples caller and noncaller buys from deciles 6 through 8 the overnight raw returns are greater than the 1 day raw returns. For such picks, prices on average fall from opening to closing on the next possible trading day after a broadcast, though the relative magnitude of the drop is small. IV. Abnormal Volume In order to measure the impact on the market of a recommendation, we use a slightly modified version of Womack s (1996) measure for Abnormal Volume (AV): the ratio of volume on date t to the average volume for the 60 previous and subsequent trading days. AV i,t ¼ V i,t ð P t 1 t 0 ¼t 60 V i,t 0 þ P tþ60 t 0 ¼tþ1 V i,t 0Þ=120

8 1120 B. Lim and J. Rosario 0.15 Noncaller Sells Table 4. Decomposed 1-day raw returns Noncaller Buys Fig. 4. Cumulative size-adjusted excess returns for noncaller picks, conditional on market capitalization Note: The horizontal lines indicate the cumulative excess return as of the market close on the date of the pick. Figure 5 graphs AV, conditional on the pick type. Day 0 refers to the date of the broadcast, with markets having already closed before Cramer makes his recommendations. Day 1 represents the first day a viewer can trade on the open market after a broadcast. For both caller and noncaller sells across all market cap groups, AV peaks on day 0, the trading day which has just concluded as Mad Money is broadcast. Equivalently, for buys, AV peaks on day 1, the next possible trading day after the pick. In omitted results, we find similar trends obtain for abnormal (number of ) trades. Defining Cramer s effect on markets as the combination of AV and excess returns on day 1, we find several congruent results: (1) Cramer s effect is higher for noncaller picks than caller picks. Confirming our earlier intuition, viewers respond less to the recommendations Cramer makes in response to callers stock questions than to those which come from his pre-planned segments. (2) Cramer s effect higher for buys than sells. If his audience is composed primarily of personal Buys Sells 1 Days Overnight 1 Day Overnight Noncaller: December (7.22) (8.56) ( 1.83) ( 1.41) Noncaller: December (13.36) (18.64) ( 2.26) ( 2.46) Noncaller: 9 and December (11.20) (9.36) ( 2.86) ( 3.40) All noncaller picks (17.25) (18.96) ( 3.91) ( 3.80) Caller: December (4.99) (6.32) ( 3.55) ( 0.50) Caller: December (7.65) (12.18) ( 4.28) ( 1.57) Caller: 9 and December (6.29) (5.29) ( 1.86) ( 3.04) All caller picks (10.50) (11.37) ( 5.49) ( 2.38) Notes: t-statistics are given in parentheses. Returns are calculated based on purchasing the stock at the closing price of the pick date. 1 Day is the return associated with selling the stock at the closing price of the next possible trading day. Overnight is the return associated with selling the stock at the opening price of the next possible trading day. investors, Cramer s viewers may lack the means or even desire to short sell the stocks Cramer views unfavourably. Certainly, most retail internet brokers do not readily provide avenues for clients to short sell. Moreover, some investors may be wary of covering a short position, given the theoretically infinite potential for losses. (3) Cramer s effect is inversely proportional to the market capitalization of the stock. This result may be considered fairly intuitive, as the average trading volume typically increases with a stock s market capitalization. It may be the case that on average the same number of viewers respond to a recommendation regardless of the stock s market cap; given that, one would expect the relative impact on smaller cap stocks to be greater than that on larger cap stocks. Another possibility is that viewers simply respond less to recommendations on large cap stocks, about which they may already have considerable information. An endorsement of a relatively unknown stock like American Shared Hospital Services (AMS) may spur more viewers to place orders than an endorsement of, say, Proctor & Gamble (PG).

9 Mad Money 1121 AV AV AV caller Sells AV noncaller Sells AV AV AV caller Buys AV noncaller Buys Fig. 5. AV: Day 0 refers to the date of the pick, with markets closing before the show airs Significance testing In order to test the significance of the AV statistics, we construct empirical distributions across the aggregated NYSE, NASDAQ and AMEX exchanges. We compute AV statistics for every stock on every trading day between 1 January 2005 and 31 December These statistics are separated by decile: for each decile, we have an empirical Cumulative Distribution Function (CDF) to compare. We define. d(i) ¼ decile of stock i. F d(i) ¼ empirical CDF of AV for decile d(i) We calculate critical values x d, such that F d (x d, ) ¼ 1. Details of the distributions are presented in the Appendix. For each pick we test whether the AV on the next possible trading day (day 1) exceeds the critical values for the pick s decile. Table 5 presents the results. Formalizing the results from the last subsection, the percentage of picks significant at generally (1) is higher for noncaller picks than caller picks, (2) is higher for buys than sells and (3) decreases with respect to the market cap grouping. As noted in ESW (2007), the calculated values for day 1 AV are likely understated, given that the data only capture trades occurring when markets are open. Presumably, there are a number of trades occurring in the afterhours trading during and subsequent to a broadcast. V. Long-Term Returns In order to estimate Cramer s forecasting ability, we calculate the long-term returns associated with his recommendations. Figure 6 plots the results, separated by pick type and market capitalization group. Detailed returns are presented in the Appendix. As was the case with shorter horizons, Cramer s small-cap picks are the most accurate over longer holding periods: subsequent to the pick date, smallcap sells trend downward while buys trend upward over the next 6 months. There is considerably volatility in these returns, owing partly to the individual volatilities of the underlying stocks and partly to the relatively small number of small-cap picks. Mid-cap picks all trend downward after the pick date, regardless of the pick type or recommendation.

10 1122 B. Lim and J. Rosario Table 5. AV significance testing Sells Buys Small cap Mid cap Large cap All Small cap Mid cap Large cap All Caller Picks Significant at 5% (10%) (8%) (8%) (8%) (13%) (7%) (7%) (8%) Significant at 1% (2%) (2%) (2%) (2%) (4%) (3%) (1%) (2%) Noncaller Picks Significant at 5% (24%) (16%) (17%) (17%) (59%) (44%) (14%) (23%) Significant at 1% (12%) (1%) (1%) (1%) (42%) (22%) (1%) (4%) Note: Percentage of picks AV significant at in parentheses Noncaller Sells 0.15 Noncaller Buys Trading days relative to pick date Trading days relative to pick date 0.15 Caller Sells 0.15 Caller Buys Trading days relative to pick date Trading days relative to pick date Fig. 6. Long-term cumulative size-adjusted excess returns, conditional on market capitalization Note: The horizontal lines indicate the cumulative excess return as of the market close on the date of the pick. That is, both Cramer and his callers are selecting mid-cap stocks which underperform the index returns for their corresponding deciles. Large-cap stocks appear largely flat after either a Buy or Sell recommendation. We attribute this trend primarily to the sheer number of picks. With several thousand large-cap pick-returns represented (and no transaction fees considered), the average of these returns would be expected to remain close to the corresponding index returns.

11 Mad Money 1123 VI. Conclusion Measuring Cramer s impact on equity markets by the combination of next-day returns and AV, we emphasize three general trends: his impact is greater for noncaller picks than caller picks, his impact is greater for smaller cap stocks and his impact is greater for buys than sells. Evidence of Cramer s forecasting ability is favourable. Mid- and large-cap post-pick excess returns are generally of the correct sign, though the magnitude of these returns is relatively small. Where Cramer displays the most ability is with small-cap stocks, in both his caller and noncaller picks. Somewhat curiously, conditional on purchasing a stock at the nextday opening price which, given the negligible differences between overnight and next-day returns, is essentially identical to the next-day closing price the highest returns are associated with long positions in small-cap noncaller sells. Long-term post-pick excess returns on noncaller picks have the proper sign and are significant but are on average only about 1% above or below their benchmarks. The long-term post-pick excess returns for caller picks are all negative, which, coupled with the positive pre-pick excess returns, suggest that callers are inquiring about not only hot stocks but also overvalued ones. Caveats Any conclusions to be derived from the preceding analysis of Cramer s stock picking ability must be considered incomplete, as Cramer s recommendations are generally too nuanced to be captured by the simple strategy implied by our examination. Like most professional stock pickers, Cramer consistently advocates an active trading style, sometimes indicating price targets, and the simple buy-and-hold strategies in this article are unable to capture the trading strategies presented on his programme with any precision. Unlike most professional stock pickers, however, Cramer does not have the luxury of selectivity. Whereas his financial sector counterpart may be content to recommend a portfolio with a relatively small number of stocks, Cramer must consistently generate new picks in order to remain relevant. Even if it were the case that Cramer s preferred portfolio consisted of, say, 50 stocks, the laws of television dictate that he must advocate hundreds more on his programme. In this sense, one must handicap Cramer s forecasting ability with respect to similar studies of mutual- and hedge-fund managers. References Barber, B. and Loeffler, D. (1993) The dartboard column: second hand information and price pressure, Journal of Financial and Quantitative Analysis, 28, Barber, B. M. and Odean, T. (2006) All that glitters: the effect of attention and news on the buying behavior of individual and institutional investors, Review of Financial Studies, 21, Busse, J. A. and Green, T. C. (2002) Market efficiency in real time, Journal of Financial Economics, 65, Engleberg, J., Sasseville, C. and Williams, J. (ESW) (2007) Attention and asset prices: the case of Mad Money, Working Paper, Kellogg School of Management. Greene, J. and Scott, S. (1999) Liquidity provision and noise trading: evidence from the investment dartboard column, Journal of Finance, 54, Keasler, T. R. and McNeil, C. R. (2010) Mad Money stock recommendations: market reaction and performance, Journal of Economics and Finance, 34, Liang, B. (1999) Price pressure: evidence from the Dartboard column, Journal of Business, 72, Liu, P., Smith, S. D. and Syed, A. A. (1990) Stock price reactions to the Wall Street Journal s securities recommendations, Journal of Financial and Quantitative Analysis, 25, Lloyd-Davies, P. and Canes, M. (1978) Stock prices and the publication of second-hand information, Journal of Business, 51, Womack, K. (1996) Do brokerage analysts recommendations have investment value?, Journal of Finance, 51,

12 1124 B. Lim and J. Rosario Appendix Table A1. AV empirical CDF Decile Observations X decile,5% X decile,1% Notes: For each decile, we calculate the empirical CDF for the distribution of abnormal volume observations. X decile, is a critical value such that F decile (X decile, ) ¼ 1.

13 Copyright of Applied Financial Economics is the property of Routledge and its content may not be copied or ed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or articles for individual use.