* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Correlated Trading and Returns
Foreign exchange market wikipedia , lookup
Financial crisis wikipedia , lookup
Investment fund wikipedia , lookup
Insider trading wikipedia , lookup
Socially responsible investing wikipedia , lookup
Commodity market wikipedia , lookup
Technical analysis wikipedia , lookup
Private equity secondary market wikipedia , lookup
Private money investing wikipedia , lookup
Futures exchange wikipedia , lookup
Hedge (finance) wikipedia , lookup
Stock exchange wikipedia , lookup
Short (finance) wikipedia , lookup
Securities fraud wikipedia , lookup
Stock market wikipedia , lookup
High-frequency trading wikipedia , lookup
Efficient-market hypothesis wikipedia , lookup
Market sentiment wikipedia , lookup
Algorithmic trading wikipedia , lookup
Trading room wikipedia , lookup
Stock selection criterion wikipedia , lookup
DNB Working Paper No. 72/December 2005 Daniel Dorn, Gur Huberman and Paul Sengmueller DNB W O R K I N G P A P E R Correlated trading and returns De Nederlandsche Bank Correlated trading and returns Daniel Dorn, Gur Huberman and Paul Sengmueller * * Views expressed are those of the individual authors and do not necessarily reflect official positions of De Nederlandsche Bank. Working Paper No. 072/2005 December 2005 De Nederlandsche Bank NV P.O. Box 98 1000 AB AMSTERDAM The Netherlands Correlated trading and returns ∗ Daniel Dorn† Gur Huberman‡ Paul Sengmueller§ Abstract Retail clients at a major German discount broker trade in tandem – they tend to be on the same side of the market in a given stock during a given day, week, month, and quarter. Neither aggregate liquidity effects nor short sale constraints fully explain this behavior. The systematic execution of limit orders, coordinated through price movements or the correlated trading of other investors who pick off retail limit orders, do not fully explain the observed comovement either. Rather, tandem trading appears to be mostly due to investors placing similar speculative bets. Correlated speculative trades perturb markets enough to make returns predictable over a short horizon. Correlated limit orders also predict subsequent returns, but for a different reason: limit order traders are compensated for accommodating other traders’ temporary liquidity demands. KEYWORDS: Retail investor trading, speculative trading, limit orders JEL codes: G11, G12 ∗ We thank Maurice Bun, Larry Glosten, Will Goetzmann, Charlie Himmelberg, Anne Jones, Frank de Jong, Alexander Ljungqvist, Suresh Sundaresan, and seminar participants at the Helsinki School of Economics for their comments. Daniel Dorn acknowledges financial support from a Dean’s research grant. † LeBow College of Business, Drexel University; 208 Academic Building; 101 North 33rd Street; Philadelphia, PA 19146; Email: [email protected] ‡ Graduate School of Business, Columbia University; 807 Uris Hall; 3022 Broadway; New York, NY 10027; Email: [email protected] § Finance Group, FEE, Universiteit van Amsterdam and De Nederlandsche Bank; Roetersstaat 11; 1018 WB Amsterdam; The Netherlands; Email: [email protected] Across all market participants, trades net to zero in each stock; there is a buyer for every seller. Yet, subgroups of investors may be net buyers or sellers in a given stock during a given period. The subgroup studied in this paper is a sample of more than 37,000 retail clients at one of the three largest German (and European) discount brokers, i.e., brokers that do not give investment advice. The clients’ complete daily transaction records, available from February 1998 to May 2000, allow us to address three related questions: First, do retail investors trade more similarly than we would expect them to merely by chance? Second, what coordinates retail trades, or, put differently, why do retail investors act similarly? Third, how does correlated retail trading affect price formation? We are not the first to consider these questions. Barber, Odean, and Zhu (2003) report that clients at a US discount broker engage in correlated trading at a monthly frequency and that past returns and trading volume help coordinate retail trades. Using the same data, Kumar and Lee (2005) report that retail trading imbalances help explain monthly stock return variation in addition to what can be explained with commonly used empirical asset pricing factors. Both studies, as well as contemporaneous work on the relation between retail trading and price formation (Kaniel, Saar, and Titman (2004) and Andrade, Chang, and Seasholes (2004)), treat retail trades as homogenous. What distinguishes our study is that we examine different types of retail trades. We distinguish between speculative and other trades, and between market and executed limit orders. These distinctions are crucial; our answers to the questions outlined above vary depending on the types of trades considered. Moreover, since our focus is mainly on the daily and weekly frequencies, we are able not only to show that correlated speculative market trades are contemporaneously correlated with returns, but also that they lead returns. Retail investors in our sample exhibit a stronger tendency to trade in tandem than the 2 institutional investors studied in earlier papers (see Lakonishok, Shleifer, and Vishny (1992) (LSV), Wermers (1999), and Wylie (2005)). In a typical stock and quarter, 57% of the investors are on the same side of the market (where one would expect 50% of the funds to be on the same side of the market simply by chance). In contrast, LSV report that only 50.1% of the US pension funds in their sample are on the same side of the market in a typical stock and quarter. Correlated trading among retail investors persists at the daily frequency with 54.3% of the investors moving in the same direction. In part, the observed comovement is due to investors being unable to short stocks. Wylie (2005) points out that investors subject to short-sale constraints will end up on the same side of the market even if their trades are otherwise uncorrelated. Such constraints explain between one fifth and one half of the comovement observed in our sample. The correlated execution of limit orders is another reason why retail investors end up on the same side of the market; for example, a price jump can trigger the execution of limit sell orders that were submitted days or even weeks apart. Indeed, we find that 56.2% of the limit order traders move in the same direction in a typical stock and day whereas only 53.9% of the market order traders do.1 However, even after controlling for the effects of short-sale constraints and limit orders, retail investors act more similarly than one would expect if they just traded randomly. Our tandem trading results contribute threefold to the existing literature on correlated trading among individual investors in the US and Australia2 : they demonstrate that correlated trading occurs at higher frequencies than previously reported, they explicitly acknowledge and quantify the roles of short-sale constraints and limit orders, and they document 1 Barber, Odean, and Zhu (2003) proxy for limit order effects by filtering out buys on days with negative returns and sells on days with positive returns. They report that their results are unchanged for the reduced sample of transactions. 2 See Barber, Odean, and Zhu (2003), Kumar and Lee (2005), and Jackson (2003). 3 correlated trading for yet another country. Why do individual investors move together? Answering this question requires understanding why individuals actively trade stocks and voluntarily take on idiosyncratic risk and high transaction costs. Knowing the investor’s identity, characteristics of his trades, and his stock, bond, fund, and options holdings and trades helps us to classify orders as likely non-speculative, i.e., driven by savings, dissavings, or risk sharing motives, and as likely speculative, i.e., driven by perceived information about the future stock price. For example, the transaction records contain a variable that identifies the order as part of an automatic investment plan through which retail investors can gradually build or reduce positions in individual stocks and mutual funds at predetermined dates (similar to ShareBuilder in the US); such plan trades are likely driven by savings or dissavings motives and are thus classified as non-speculative. Trading in assets other than common stocks also helps to assign trading motives; for example, if a stock sale is followed by the purchase of a stock mutual fund, we classify the sale and the purchase as diversification-driven and thus non-speculative. Based on our definition of speculative trading, which extends the definition of Barber and Odean (2002), we estimate that more than half of the observed stock trading volume is due to speculative trading. Remarkably, the observed correlated trading among retail investors is primarily driven by speculative trades – depending on the observation frequency, the observed comovement among speculative market order trades is 20%-50% higher than among non-speculative market order trades. In other words, retail investors move together primarily because they tend to place similar speculative bets, not because their non-speculative trades are coordinated through, say, automatic investment plans or shared return experiences. Interestingly, investor location has little impact on speculative trading behavior (contrary to Feng and Seasholes (2004)); local speculators tend to be on the same side of the market as remote 4 speculators, suggesting that the two investor types react similarly to public signals. Ours is one of the first studies to examine what motivates retail trades and to shed light on what accounts for the observed correlated trading. The results complement those of Grinblatt and Keloharju (2001) and Barber, Odean, and Zhu (2003) who examine the common stock transactions of Finnish and US individuals. These papers rely mostly on historical stock price patterns to infer what makes people trade. In contrast, we use trades in other securities as well as additional trade characteristics to classify stock trades as speculative versus non-speculative. Correlated decisions to trade individual stocks have implications not only for individual investor welfare (see, e.g., Barber and Odean (2000)), but potentially for prices as well. Individual investors could play a role in price formation because their orders, taken together, demand liquidity from or supply liquidity to other market participants. Individual investors might move prices because their speculative trades reveal information about future prices (see Glosten and Milgrom (1985) and Kyle (1985)). And even if investors traded on signals unrelated to fundamental values as conjectured in models of noise trading and style investing, their trading could move prices as long as other investors are constrained from betting against them (see, e.g., De Long, Shleifer, Summers, and Waldmann (1990a), Shleifer and Vishny (1997), or Barberis and Shleifer (2003)). In our sample, days and weeks with heavy speculative market order buying are associated with high returns, while speculative market order selling is associated with low returns; the tercile of stocks most aggressively bought by speculators outperforms the tercile of stocks most heavily sold by 1.8% during the day and by 4.7% during the week of portfolio formation. More remarkably, the observed speculative trading leads returns; the speculative buy portfolio continues to outperform the speculative sell portfolio by 0.56% during the day and 5 by 0.6% during the week after portfolio formation. This appears to be inconsistent with speculators providing liquidity to other market participants as one would expect returns to revert to their pre-price pressure level. One interpretation of our results is that speculative traders, due to a persistence not anticipated or accommodated by other market participants, perturb markets enough to make returns predictable over a short horizon. As a by-product of our analysis, we document that speculators behave as positive feedback traders, i.e., they tend to buy good past performers and sell poor past performers. These results stand in stark contrast to situations where individuals trade together via limit orders, presumably taking a less aggressive stance when compared with correlated market orders. Limit trades appear to be negative feedback trades. Moreover, limit order imbalances are strongly negatively correlated with contemporaneous returns both at the daily and weekly frequency. In large part, the correlations between limit order imbalances and past as well as contemporaneous returns are due to the strong mechanical relation between price movements and limit order execution (see also Linnainmaa (2003)). However, limit order imbalances are also strongly positively correlated with future returns which is consistent with the hypothesis that limit order traders are compensated for accommodating temporary liquidity demands of other investors. Being able to proxy for the motivation of trading and incorporating information about the microstructure of order execution turns out to be crucial for understanding the relation between individual investor trading and returns. Our paper distinguishes between limit orders and market orders, and market orders further between speculative orders and nonspeculative orders. For example, combining market and limit orders leads us to classify retail investors as contrarian, similar to prior and contemporaneous papers studying individual investor behavior (see Grinblatt and Keloharju (2000), Nofsinger and Sias (1999), 6 Griffin, Harris, and Topaloglu (2003), Barber, Odean, and Zhu (2003), Kaniel, Saar, and Titman (2004), Goetzmann and Massa (2002), and Jackson (2003) for evidence on Finnish, US, and Australian retail investors). However, the mechanical correlation between limit order execution and returns masks the previously undocumented relations between speculative trading – arguably a cleaner measure of retail opinion than limit trading – and past as well as contemporaneous returns. The predictive relation between speculative trading and future returns suggests that retail investors affect price formation beyond supplying liquidity through limit orders. This high-frequency evidence complements Kumar and Lee (2005) who report that retail trading imbalances help explain monthly stock return variation in addition to what can be explained with commonly used empirical asset pricing factors. The remainder of the paper proceeds as follows: Next is a description of the transaction records. Section Three examines comovement among the sample investors. Section Four relates correlated speculative trading to returns, and Section Five concludes. I Data The analysis in this paper draws on complete daily transaction records for a sample of more than 37,000 clients at one of the three largest German discount brokers between February 1998 and May 2000. The client sample was drawn randomly from the entire active and former client population in January 1999 and represents roughly half of the existing client population at that time – hence, the sample is free of survivorship bias. The broker is labelled as a discount broker because no investment advice is given. This is important because it rules out correlated trading due to broker recommendations.3 In June 2000, at the end of 3 Initial public offerings (IPOs) are an exception. During the sample period, IPOs are prominently displayed on the broker’s web site. Moreover, the broker acts as an underwriter in several IPOs which could be interpreted as an endorsement even if no explicit recommendation is given. 7 our sample period, there were almost 1.5 million retail accounts at the five largest German online or discount brokers (Van Steenis and Ossig (2000)) – a sizable number, given that the total number of German investors with exposure to individual stocks at the end of 2000 was estimated to be 6.2 million (see Deutsches Aktieninstitut (2001)). The records are complete in that they contain the opening position as well as all transactions of a client from the account opening date until the account closing date or May 31, 2000 whichever comes earlier. This allows us to reconstitute the entire portfolio of each client at the end of each day during the sample period. The typical transaction record consists of a unique identification number, an account number which identifies the client, a transaction date, a buy/sell indicator, order type (indicating whether an order is a limit or a market order), order channel (indicating whether the order was placed, e.g., by phone, internet, or as part of an automatic savings plan), an exchange indicator (identifying the exchange on which the order is placed), the type of asset traded (e.g., common stock), a security identification code, the number of shares traded, the gross transaction value, and the transaction fees. In principle, brokerage clients can trade all the bonds, stocks, and options listed on German exchanges, as well as all the mutual funds registered in Germany by domestic and foreign issuers. The database also identifies short-term deposits, money market funds, and the nature of mutual fund investments (e.g., stock or bond, sector, geographic focus), but contains no direct information on client cash balances. To examine correlated trading, we focus on transactions in domestic stocks though transactions in other assets are considered to infer trading motives. Table I summarizes the client portfolios and trading activity during the sample period. At the end of the sample period, the typical client portfolio is worth DEM 45,000 (USD 23,000 at the average DEM/USD rate of approximately 2 DEM per USD during 8 the sample period) and consists entirely of individual stocks and stock mutual funds. The Herfindahl-Hirschmann Index (HHI) of the typical equity portfolio is 13.5%, i.e., clients hold the equivalent of seven stocks equally weighted.4 In aggregate, the brokerage positions are worth DEM 4.6 billion, almost 90% of which is in equities. The aggregate trading volume, the sum of absolute values of purchases and sales during the sample period, is DEM 18.3 billion. Most of the trading occurs in individual stocks and options. ************************* TABLE I HERE ************************* We use Datastream to retrieve daily returns adjusted for splits and dividends for all German exchange-listed stocks, all registered mutual funds, and the DAX 100 and the Neuer Markt 50 (Nemax 50) indices5 . The broker provides its own database for bonds and options prices. As a robustness check, we examine Neuer Markt stocks separately since highquality daily share turnover and intra-day prices are available for these stocks through the Karlsruher Kapitalmarktdatenbank (a database, maintained by the University of Karlsruhe, which makes German Stock Exchange data available at cost for academic purposes). 4 The HHI is defined as the sum of squared portfolio weights. A portfolio consisting of n stocks equally 2 weighted would have an HHI of n · n1 . Note that we assume stock mutual funds to consist of one hundred equally weighted positions that do not overlap with other holdings of the investor. That is, the HHI of portfolio of an investor holding one mutual fund is 1% and that of an investor splitting his money equally between two mutual funds is 0.5%. 5 The Neuer Markt was the market segment for growth and technology stocks of the Frankfurt Stock Exchange. It was founded in March 1997 and closed six years later. 9 II A Correlated trading Baseline results To detect herding among US pension funds, Lakonishok, Shleifer, and Vishny (1992) (LSV) develop a herding measure which has been subsequently used to assess the correlated trading behavior of, e.g., US mutual funds (Grinblatt, Titman, and Wermers (1995) and Wermers (1999)), UK mutual funds (Wylie (2005)), different groups of Finnish investors (Kyrolainen and Perttunen (2002)), and a sample of US discount brokerage clients (Barber, Odean, and Zhu (2003)). The LSV measure, HLSV , is defined as HLSV = PT 1 t=1 Nt XX brjt − brtLSV − Et brjt − brtLSV t (1) j where Nt is the number of stocks traded during period t, the buyers ratio brjt is the number of net buyers of stock j during period t divided by the number of active traders of j during t, and the period-average buyers ratio brtLSV is the number of net buyers aggregated across all stocks during t divided by the number of all active traders during t. Subtracting the period-average buyers ratio controls for aggregate shifts into or out of stocks, e.g., in response to liquidity shocks.6 ************************* TABLE II HERE ************************* Table II shows the baseline results for the sample of German brokerage clients, consid6 To get a test statistic that is zero under the null hypothesis of no correlated trading, LSV subtract the expected value of the difference between buyers ratio and period-average buyers ratio, I jt X Et brjt − brtLSV ≡ k=0 Ijt k brtLSV 10 k 1 − brtLSV Ijt −k k LSV − br t Ijt ering daily, weekly, monthly, and quarterly horizons during the entire sample period 2/1/98 - 5/31/00. We only consider secondary market purchases of German stocks. Note that LSV measures are only calculated for stock-periods with at least two active traders. There is correlated trading across all horizons and the correlation appears to increase with longer observation intervals. At a quarterly horizon, for example, the average LSV measure is 8.3% and the median LSV measure is 7%; assuming that the average fraction of position changes that are increases is one half, 57% of the sample brokerage clients are on the same side of the market in a typical stock quarter (see Panel D, Column (1), of Table II) - by contrast, LSV report that only 50.1% of their sample pension fund managers trade in the same direction in a typical stock quarter. Remarkably, and also in contrast to LSV, the LSV measure is higher when calculated across active trading periods: across stock-quarters that belong to the top trading quartile7 , the average is 9.4% (see Panel D, Columns (2) and (3)). When measured over shorter intervals, the average LSV measure declines monotonically from 6.4% at a monthly horizon to 4.8% at a daily horizon (see Panels A-C, Column (1)). Again, the LSV measure is higher when calculated across active trading periods. On stock-days in the top trading activity quartile, 57.1% of the clients are on the same side of market as opposed to 54% in the bottom three quartiles (see Panel A, Columns (2) and (3)). The results are similar when the period-average buyers ratio is estimated as the ratio of DEM purchases to DEM trading volume, i.e., using trading volume rather than traders. We verify that the observed comovement is not due to instances of mechanically correlated trading such as secondary market purchases of new stock issues (the results are virtually unchanged when excluding the first six months of IPO trades), exchanges of one stock for another at the conclusion of a merger or an acquisition or the distribution of bonus shares 7 For each stock, we select the quartile of observations with the broadest participation by the sample investors as measured by the number of net traders in a given period. As a consequence, each stock is represented roughly equally in the top trading quartile. 11 (the database identifies such transactions separately and we omit them), or stock repurchases. All reported LSV measures are highly statistically significant at conventional levels assuming, as LSV and subsequent papers do, that observations are independent across time and across stocks. These assumptions are not quite innocuous: if the investors’ tendency to buy or sell is serially correlated – suggested by the increase in correlated trading at longer intervals – observations will not be independent across time. Even if we assume that observations are only independent across stocks, however, the LSV measures continue to be strongly significant (see Column (4) of Table II). Another possibility is that the observed comovement across stocks is a by-product of the investors moving together into and out of industries. If so, comovement measured across industries should be stronger than across individual stocks. Column (5) of Table II reports the LSV measures calculated for Level 6 (most detailed) Datastream Industry Classifications; the measures for coarser Datastream classifications are similar. The industry LSV measures are all positive and significant and increase as the observation period lengthens from days to quarters; their magnitude, however, is lower than that for individual stocks. LSV and subsequent papers control for aggregate shifts into and out of the sample stocks by subtracting the period-average buyers ratio across all sample stocks – normal net buying activity – from the individual stock buyers ratios. The period-average buyers ratio across both stocks and stock mutual funds, instead of just individual stocks, may be a better measure of the normal net buying activity – e.g., liquidity-driven investors may prefer trading mutual funds to trading individual stocks to avoid trading with informed investors as pointed out by Subrahmanyam (1991) and Gorton and Pennacchi (1993). Moreover, a buyers ratio aggregated across all equity securities that can be traded through the broker should give us a better idea of the money added to or withdrawn from the equity portion of the accounts. 12 Column (6) of Table II shows the LSV measures for German stock trades using the more comprehensive period-average buyers ratio as a control for liquidity shifts. Average LSV measures are considerably higher; at a quarterly horizon, for example, the LSV measure averages 10.3% when the average buyers ratio is calculated across all equity securities versus 8.3% when the average buyers ratio is calculated across individual German stocks only. The period-average buyers ratio used in the baseline calculations above thus appears to produce a conservative estimate of correlated trading due to reasons other than aggregate liquidity effects. B Robustness checks B.1 Short-sale constraints The standard interpretation of the LSV statistics entails the assumption that a sale of a stock is just as likely as its purchase. For this to be true, short sales must be allowed and routinely executed. In many contexts, however, investors are legally barred from selling short (e.g., UK fund managers (Wylie (2005))) or promise not to sell short (e.g., long-only institutional money managers), or simply follow long-only strategies (Barber, Odean, and Zhu (2003)). Moreover, some stocks are hard to short, especially for retail investors. During our sample period, none of the major German retail brokers allows its clients to short stocks.8 When investors are short-sale constrained, investors with prior holdings in a stock can choose between an additional purchase, no action, and a sale whereas investors without prior holdings can only choose between an initial purchase and no action. Wylie (2005) reports 8 The brokers point to the unlimited downside risk associated with short selling and the resulting threat of legal action by retail clients; according to paragraph 37d of the German Securities Trading Law (Wertpapierhandelsgesetz WpHG), a broker can be liable for damages arising from certain transactions if the broker failed to fully disclose the risks associated with the transaction – see the edition of http : //www.f az.net, of May 24, 2002, ”Leerverkäufe – leichter gesagt als getan” (Short-selling - easier said than done), last viewed August 15, 2002. 13 that, even under the null of no herding, the LSV measure becomes significantly positive when investors face short-sale constraints. Intuitively, if few of the sample investors hold shares in a given stock, random initial purchases of that stock by other sample investors might be classified as correlated purchases. To assess the potential bias induced by short-sale constraints, we adopt a procedure similar to that in Wylie (2005); the procedure is described in Appendix A. Table III summarizes the simulation results. In the presence of short-sale constraints, LSV measures are upwardly biased. Even if trading were uncorrelated, one should expect to observe LSV measures of around 2% across all horizons which is significantly different from zero (by comparison, Wylie (2005) finds a bias of 1.5% for a semi-annual horizon). In other words, one would expect to see 52% of the sample investors to be on the same side of the market in a typical stock-period even if their trading were not coordinated other than by short-sale constraints. ************************* TABLE III HERE ************************ In short, short-sale constraints help coordinate trading; the relative magnitude of correlated trading that can be attributed to short-sale constraints is similar to that found by Wylie (2005). However, short-sale constraints by themselves capture neither the full extent of the observed comovement nor the concentration of trading activity in relatively few stocks especially at short horizons. 14 B.2 Limit orders Could high LSV measures simply reflect the coordinated execution of stale or otherwise unrelated limit orders? During a high-return day, for example, the execution of limit sell orders will be coordinated by the rising stock price. Alternatively, the observed comovement could be a by-product of other traders picking off the sample investors’ limit orders, i.e., the sample investors might be providing liquidity to the market (see Kaniel, Saar, and Titman (2004)). The data allow us to distinguish between executed market and limit orders; limit orders can be classified as regular, stop buy, and stop loss orders. Although there is no information about the limit level, the data base identifies the limit expiration date as either the day of order submission, the last trading day of the current month, the last trading day of the next month, or “good until cancelled.” The majority of executed limit orders – 55% – have a limit expiration date beyond the day of order submission. Presumably, the proportion of unmonitored limit orders is higher among orders with longer limit expiry dates. To get a conservative estimate of how limit orders affect trading correlations, we calculate LSV measures separately for market orders and for limit orders - regular limit, stop buy, and stop-loss orders. The results, reported in Table IV, indicate that the LSV measures based on market orders are about 10% less than the LSV measures based on all orders. Moreover, LSV measures based on market orders are also smaller than the LSV measures based solely on limit orders; the difference between market order comovement and limit order comovement depends on the observation frequency. ************************* TABLE IV HERE ************************* At a quarterly frequency, there is little difference between market-order based LSV and limit-order based LSV. By contrast, at the daily frequency, the distortion of the LSV mea15 sure due to the correlated execution of limit orders is considerable – stock prices both rise and fall during a quarter, triggering the execution of limit buy and limit sell orders, whereas on a given day, stock prices tend to move in one direction if only because of a relatively small number of price fixings, triggering the execution of either limit buy or limit sell orders. On stock-days belonging to the top trading quartile, for example, the LSV measure based on limit orders averages 8.3% versus 6.2% based on market orders. The impact of short-sale constraints is similar to that presented in Section B.1. Shortsale constraints explain between one fifth of the comovement in market orders at a quarterly horizon and slightly more than half of the observed comovement at a daily horizon. All actual LSV measures reported in Table IV remain significantly different from the simulated averages – assuming that the sample investors trade randomly subject to short-sale constraints – as reported in Table III. Thus, the observed comovement is not merely a by-product of shortsale constraints or the coordinated execution of limit orders or both. C Why do retail investors move together? People trade for different non-speculative reasons: they might purchase stocks to save for retirement and sell stocks to make the down payment for a house, to diversify risk, to rebalance their portfolio, or to lower their tax bill. In addition, investors trade speculatively; they trade to exploit perceived information about stocks. Similar motives for trading might coordinate trades in addition to the reasons outlined in Sections B.1 and B.2. If, for example, many clients invest a fraction of their monthly paycheck through the brokerage account, we will observe correlated purchases around the payday. The taxation of capital gains might coordinate the selling of losers towards the end of the year as investors try to offset capital gains realized during the year (see Odean (1998)). 16 Although it is impossible to divine the exact motivation behind a trade, the transactions data allow us to divide trades into those undertaken for mostly speculative reasons and those undertaken for mostly non-speculative reasons such as savings, liquidity, or portfolio rebalancing within an asset class or across asset classes. Note that there is little reason to suspect much tax-motivated trading in the sample. Capital gains in Germany are not taxable unless they are realized within a certain period known as the speculation period (six months before 1999, twelve months from 1999 on). Unlike in the US, however, German banks and brokers are not required to report client transactions to the German IRS which makes it impossible to adequately enforce this tax law; in 2004, the German Supreme Court (Bundesverfassungsgericht) ruled the speculation tax – dubbed the “dunce tax” in honor of the few investors who faithfully report speculation gains – unconstitutional, upholding an earlier judgment by the German Finance Court (Bundesfinanzhof). C.1 Speculative versus non-speculative trading Using individual-level transactions in common stocks from a US discount broker, Barber and Odean (2002) define speculative trades as “all profitable sales of complete positions that are followed by a purchase within three weeks and all purchases made within three weeks of a speculative sale.” (p. 475) They require that complete positions be sold to filter out rebalancing sales, profitable positions be sold to filter out tax-loss sales, and that sales are closely followed by purchases to filter out liquidity sales. Our baseline definition of speculative trading corresponds to that of Barber and Odean, with two modifications. First, we include stock sales for a loss as well – as argued above, such sales are unlikely to be tax-motivated. Second, our transaction records indicate whether a trade is part of an automatic savings or withdrawal plan that exist for dozens of individ17 ual stocks and mutual funds. Clients who sign up for such a plan can contribute between DEM 100 and DEM 5,000 to their chosen fund or stock at four pre-determined dates each month (similar to ShareBuilder in the US); plan withdrawals, which are extremely rare, work similarly. Plan transactions are likely driven by savings or dissavings motives and are thus classified as non-speculative. Table V reports the breakdown of transactions, by number and DEM volume, into speculative and non-speculative purchases and sales. Roughly three out of five purchases and sales are classified as speculative. Speculative purchases represent almost 70% of total purchase volume in DEM; this is partly due to investment plan transactions which are relatively small and automatically classified as non-speculative. ************************* TABLE V HERE ************************* To check the robustness of this classification and to get a better idea of what motivates individual stock trades, we develop a second definition of speculative trading using the investors’ transactions in money market securities, bonds, bond funds, stock funds, index certificates, and options. We classify an individual stock sale as speculative if the sale is for the entire position and, within three weeks, the sale is followed by the purchase of another individual stock or the purchase of an option that does not hedge an underlying position9 (all calls and some puts). We classify sales of individual stocks as non-speculative if the investor sells within an automatic withdrawal plan, only sells part of her position (rebalancing-driven sales, see Barber and Odean (2002)), or, within three weeks, the sale 9 Since the sample investors cannot write options, they have to buy puts to go short. A put hedges an existing position if the underlying substantially overlaps with the investor’s portfolio at the time of the purchase. We consider the following substantial overlaps: 1. if the investor buys a put on an individual stock, she must be long the stock, 2. if the investor buys a put on an industry basket, she must be long stocks in the same industry, 3. if the investor buys a put on a foreign currency, she must be long the foreign currency, and 4. if the investor buys a put on an index, she must either own index components or similar securities (e.g., if an investor buys a put on the German market index DAX, the purchase is classified as hedging driven if the investor owns German stocks). 18 is followed by either a money market fund purchase or no purchase at all (liquidity- or dissavings-driven sales), a stock fund purchase (diversification-driven sales), a bond or bond fund purchase (allocation-driven sales), or a put option purchase if the put option hedges an existing position (hedging-driven sales). Note that this classification scheme is exhaustive but does result in some conflicts. For example, an individual stock sale might be followed within three weeks by the purchase of a stock fund (leading us to classify the sale as diversification-driven and thus non-speculative) and the purchase of an individual stock (leading us to classify the sale as speculative). In these ambiguous cases, we classify the sale itself as non-speculative, but record it as both speculative and non-speculative to classify subsequent purchases. We classify purchases of individual stocks as speculative if they follow within three weeks of a speculative individual stock sale, a stock fund sale, or an options sale. We classify purchases of individual stocks as non-speculative if they occur within an automatic savings plan, or they are preceded, during the prior three weeks, by no sale at all (savings-driven purchases), a money market fund sale (liquidity-driven purchases), an individual bond or bond fund sale (allocation-driven purchases), or a rebalancing sale (rebalancing-driven purchases). In case of a classification conflict, e.g., if a purchase is preceded by both an individual stock sale and a bond sale, we use the most recent sale to classify the purchase (and classify the purchase as non-speculative if the competing sales happen on the same day). According to this alternative definition of speculative trading, slightly more than half of the observed trading is classified as speculative (see Table V). Again, because of smaller savings-plan purchases, the ratio of speculative purchase volume to total purchase volume (61.2%) exceeds the ratio of number of speculative purchases to total number of purchases (52.2%). Savings and rebalancing appear to be the dominant motives for non-speculative trading; there is very little trading across asset classes.10 10 The magnitude of speculative trading is fairly insensitive to different assumptions. For example, if we 19 Both definitions of speculative trading are conservative in that trades identified as speculative – typically a rapid succession of sales and purchases of entire individual stock positions that typically leave the investor’s overall exposure to equities unchanged and occur in portfolios that consist only of a handful of stocks – are almost certainly not due to savings, liquidity, or risk sharing motives. At the same time, some of the trades classified as nonspeculative by both definitions may be speculative in nature. For example, improvements in diversification brought about by “rebalancing” a winning stock position may very well be a by-product of pursuing the next speculative investment idea. The alternative definition of speculative trading is more conservative as trades between asset classes tend to be classified as non-speculative, although such trades may be motivated by market timing. Do the sample investors trade in tandem because of overlapping non-speculative trading motives or because they place similar speculative bets? One would expect non-speculative trades to be coordinated beyond the correlation implied by short-sale constraints, if only because some of the trades occur within automatic savings plans (which coordinate purchases of certain stocks on certain days) or because rebalancing trades are triggered by shared return experiences. Indeed, the LSV measures calculated for non-speculative market orders, reported in Column (1) of Table VI, are significantly positive at all horizons. Perhaps surprisingly, however, speculative market orders are more strongly correlated than nonspeculative trades at all horizons (see Column (2) of Table VI). At a quarterly horizon, for example, the LSV measure calculated for speculative trades is almost twice that for nonspeculative trades. Similar results obtain when trades are classified using the alternative extend the “turnaround” period from three to six weeks – assuming that clients have better options to get liquidity for a six-week period than by selling stocks – the fraction of individual stock trading classified as speculative climbs four percentage points. If we assumed that shifts between asset classes such as stocks and bonds are motivated by market timing and are thus speculative, the volume of speculative trading increases by two percentage points. 20 measure of speculative trading (reported in Columns (3) and (4) of Table VI). These results suggest that correlated trading among the sample investors is mainly driven by speculative trading, i.e., trading in response to perceived signals about the future price of a stock. ************************* TABLE VI HERE ************************* C.2 Speculative trading and location If there is enough noise such that prices do not fully reveal the information about the fundamental value of a stock, there is scope for trading between asymmetrically informed investors. Feng and Seasholes (2004) conjecture that local investors – those who live close to a firm’s headquarters – have more precise priors about the value of the firm than remote investors – those who live far away from the firm’s headquarters. Once public information arrives, both local and remote investors update their beliefs but remote investors update more - because of their less precise priors - leading to trade between the two groups. To check whether the observed correlated speculative trading in a firm is the net effect from trading between local and remote investors, we classify investors firm by firm into those who live closer to the firm’s headquarters than the typical holder of the firm throughout the sample period (local investors) and the remainder as remote investors.11 First, following Feng and Seasholes (2004), we ask whether the number of net local speculative buyers (local speculative buyers - local speculative sellers) and net remote speculative buyers is significantly correlated at a daily frequency. The correlation of local and remote net buyers across 11 Given latitude (lat) and longitude (lon) coordinates for respondent i and firm j, the distance between i and j is calculated as di,j = earth radius · arccos (sin(latj ) sin(lati ) + cos(latj ) cos(lati ) cos(lonj − loni )) 21 all stocks and during the period 2/1/98 - 5/31/00 is greater than 0.512 - in contrast, Feng and Seasholes (2004) document a negative correlation between the number of net buyers at different regional branches of a Chinese brokerage house. Moreover, LSV measures calculated separately for local and remote investors are similar as well. The differences between the Chinese sample and the German sample might be due to differences in information availability and the way orders are submitted. Feng and Seasholes (2004) study investors who submit orders in trading rooms whereas virtually all trades in our sample are submitted either by phone or through the broker’s web site. Through the web site of the German broker, clients have access to real-time financial news from sources such as Reuters. Our results suggest that local and remote investors do not trade on differential private information; rather, both local speculators and remote speculators appear to respond similarly to public signals. III Speculative trading and returns The previous section identifies trading due to speculative motives and shows this speculative trading to be correlated. To a large extent, individuals are on the same side of the market not because of liquidity, short-sale constraints, location, or limit orders. Rather, they act similarly because they follow specific trading ideas about specific stocks and take similar bets. Effectively, the observed individuals bet against the rest of the market. To what extent are these bets inspired by past returns? How are they correlated with contemporaneous returns? To what extent do they lead future returns? 12 The correlation between the net local buyers ratio, defined as the local buyers ratio minus the periodaverage buyers ratio, and the net remote buyers ratio is also positive with a coefficient of 0.35. 22 A Hypothesis development The starting point, or null hypothesis, is that there is no relation between the observed retail trading and future returns. The correlated trading plans of the sample investors are accommodated by other market participants without affecting prices. The null hypothesis encompasses the case of individual investors predicating their trades on past returns – return chasing or contrarian trading – without affecting prices. Theory offers several alternative hypotheses about the relation between correlated trading and returns. These hypotheses are summarized in Table VII. According to the “informed trading hypothesis,” if retail trades conveyed information about the future prices of securities in a world with private information (see for instance Kyle (1985) or Glosten and Milgrom (1985)), trade direction and contemporaneous as well as future returns should be positively correlated as the information is incorporated into prices - the correlation should be the stronger, the more aggressive informed trading.13 ************************* TABLE VII HERE ************************* The formulation of liquidity hypotheses crucially depends on whether the sample investors’ trades accommodate other traders’ demands or the sample investors’ trades need to be accommodated. The “liquidity demand hypothesis” states that retail buyers ratios are positively correlated with contemporaneous price changes and negatively correlated with subsequent price changes. In the underlying model, inventories of broker-dealers move away from desired levels as they accommodate their clients’ demands. In order to return to desired levels, the 13 At least if price impact functions are linear. For more discussion see Huberman and Stanzl (2004). 23 broker-dealers change the prices of the securities. Moreover, to the extent that their clients’ demands are temporary, inventories and bid and ask prices will revert to their earlier levels. For instance, if the clients exert buying pressure, the broker-dealers will sell to clients from their own inventories but will also raise both the bid and the ask prices, to entice clients to sell more to them (by raising the bid price) and to cause fewer purchase (by raising the ask price). As the clients respond to the prices, inventories will increase and bid and ask prices will come back down. Stoll (1978), Amihud and Mendelson (1980), and Grossman and Miller (1988) elaborate on this process. The “liquidity supply hypothesis” recognizes that retail buyers ratios far away from one half or the period-average buyers ratio could result from other traders picking off retail orders; in other words, the observed buyers ratio might be a by-product of the sample investors’ accommodating other investors’ liquidity demands (presumably via limit orders). In this case, imbalances should be negatively correlated with contemporaneous returns if retail investors are compensated for their supplying liquidity and positively correlated with future returns as the price pressure eases (see also Campbell, Grossman, and Wang (1993)). It is possible, of course, that the sample investors act as liquidity providers during certain periods and as liquidity demanders during other periods which makes it harder to reject the liquidity hypotheses. The “noise trading hypothesis” holds that retail trading might affect prices even if the underlying orders are based on signals unrelated to fundamental value and inventory concerns are absent – provided it is somehow undesirable for other investors to trade against the noise (see, e.g., De Long, Shleifer, Summers, and Waldmann (1990a), Shleifer and Vishny (1997), Hong, Lim, and Stein (1999), Barberis and Shleifer (2003)). For example, if noise traders engaged in predictable positive feedback trading, their buyers ratios could be positively correlated with contemporaneous returns (see De Long, Shleifer, Summers, and Waldmann 24 (1990b)) and negatively correlated with future returns, as the price moves back to fundamental value. Contrary to the other hypotheses, departures from fundamental value under the noise trading hypothesis can be protracted as rational investors are subject to limits of arbitrage. Theory offers little guidance as to the period over which future returns should be measured, i.e., how long it takes for private information or shocks due to liquidity or noise trader demands to be absorbed by the market. Prior empirical research has not reached a consensus, either. Chordia, Roll, and Subrahmanyam (2005) report adjustment speeds for a selection of NYSE stocks – based on market-wide imbalances of mostly large and liquid stocks – of just a few minutes, in apparent contrast to Chordia and Subrahmanyam (2004) who report that market-wide daily imbalances in a much broader sample of NYSE stocks are positively correlated with next-day returns. We thus consider multiple – daily and weekly – horizons in our empirical approach. It should be noted that the null hypothesis of no relation between trading and returns encompasses the case of a quick convergence to market efficiency – if prices quickly reflect the information conveyed by the sample orders, we will be unable to detect price effects with daily data. B Empirical evidence To examine the relation between speculative trading and returns, we sort the sample stocks every period by the observed buyers ratio based on speculative market orders. According to its buyers ratio rank, each stock is sorted into one of three portfolios of roughly equal size, i.e., similar number of stocks; for simplicity, we will refer to the three portfolios as buy, hold, and sell portfolios. 25 Table VIII reports the average buyers ratio of the three portfolios during the period of portfolio formation as well as during the periods preceding and following portfolio formation. On the day of portfolio formation, the average buyers ratios of the sell, hold, and buy portfolios are 0.13, 0.46, and 0.81, respectively (see Panel A of Table VIII, Column (0), reported together with average cutoffs on the formation day). Moreover, speculative trading appears to be persistent; stocks in the buy portfolio on the day of portfolio formation also tend to be bought on adjacent days whereas stocks in the sell portfolio tend to be sold on adjacent days. The daily sell, hold, and buy portfolios consist of ten stocks, on average. Panel B of Table VIII reports a similar pattern for the weekly frequency. The weekly sell, hold, and buy portfolios consist of 27 stocks, on average. ************************* TABLE VIII HERE ************************* To shed light on the hypotheses formulated in Section A, we calculate excess returns of the three portfolios for the portfolio formation period as well as the adjacent periods. Note that stock returns during period t are defined as the ratio of adjusted closing price for period t divided by adjusted closing price for period t − 1 (prices are adjusted for splits and dividends). The raw portfolio return is the equally-weighted average return across stocks. The excess portfolio return is the raw portfolio return minus a benchmark portfolio return; the benchmark portfolio consists of those German stocks that represent at least 0.5% of the total value of German stocks held by the brokerage clients. PF1, the sell portfolio, is the equally-weighted portfolio of the tercile of stocks most aggressively sold by speculators and PF3 is the corresponding buy portfolio. On the day of portfolio formation, there is a strong positive correlation between returns 26 and the direction of speculative trading. On average, the portfolio of stocks most aggressively bought by speculators has an excess return of 1.6% per day, whereas the portfolio of stocks most aggressively sold declines by 0.2%. We also form a zero-cost portfolio, calculated by subtracting the sell portfolio PF1 from the buy portfolio PF3. (Line 4, Table IX). On the day of portfolio formation, this portfolio has a large and statistically significant return of 1.8%. ************************* TABLE IX HERE ************************* The positive correlation between speculative trading and contemporaneous returns is inconsistent with the liquidity supply hypothesis; if speculators took the opposite side of liquidity traders who are forced to make price concessions, one would expect buyers ratios to be negatively correlated with returns. Evidence on next-day returns draws a clearer line between the remaining hypotheses, all of which are not inconsistent with the positive same-day correlation. Column (+1) of Table IX presents returns of the buy minus sell portfolio on the day following portfolio formation. Remarkably, this zero-cost portfolio has an economically and statistically significant return of 0.56%. This casts doubt on the investors’ conditioning their trades on past returns without affecting prices – intra-day return chasing can explain the same-day correlation, but not the correlation between today’s trading and tomorrow’s returns. The evidence also appears to be at odds with hypotheses predicting short-lived price pressure effects such as the liquidity demand hypothesis. That said, the evidence for serially correlated buyers ratios in Table VIII suggests persistent price pressure as a possible explanation for the observed correlation between today’s trading and tomorrow’s returns. If speculative trading exerted price pressure, then a high buyers ratio today would cause price changes today and predict a buyers ratio tomorrow. Tomorrow’s buyers ratio, in turn, would 27 affect tomorrow’s prices. Individuals’ speculative trades are not contrarian. If anything, the sample investors are momentum traders; as a group, they buy stocks that performed well during the prior trading session. Column (-1) of Table IX documents this tendency for stocks with previously high excess returns to get sorted into the speculative buy portfolio. The zero-cost portfolio that is long the stocks speculatively bought and short the stocks speculatively sold has a significantly positive return of 1.6% on the day preceding portfolio formation. To check the robustness of our results, we recalculate Table IX by 1. using an equallyweighted market portfolio as a benchmark instead of a holding-weighted brokerage portfolio (of course, the choice of market benchmark does not affect the returns of the zero-cost portfolio), 2. using the normalized order imbalance, defined as the difference between the number of shares bought and sold speculatively divided by the total number of shares outstanding, instead of the buyers ratio to form the portfolios, 3. using the alternative definition of speculative trading, and 4. using four instead of three portfolios. The results are robust to these differences in specification. Another explanation for the results could be that the brokerage clients buy stocks that exhibit return momentum over several days. To check this explanation, we sort stocks each day into momentum deciles by yesterday’s returns. Then, we construct portfolio returns for day -1, day 0, and day 1 by equally weighting the difference between the raw return and the return of the corresponding momentum decile for each stock in the portfolio. Our inferences are robust to this return correction as well. Despite the impressive returns of the zero-cost portfolio on and around the day of portfolio formation, economic significance is hard to establish at a daily horizon. Furthermore, liquidity imbalances may persist for several days. We therefore repeat the portfolio exercise 28 with weekly data. The results are shown in Table X. Despite the longer horizon – and a reduction of the number of observations from 609 to 122 – the correlations between speculative trading and past, contemporaneous, and future returns are similar to those obtained for the daily horizon. The zero-cost portfolio has a significantly positive return of 4.7% during the week of portfolio formation. Subtracting the return of the speculative sell portfolio from that of the speculative buy portfolio yields a statistically significant return of 0.6% during the week after portfolio formation. Again, the zero cost portfolio’s return of 2.1% during the period prior to portfolio formation is consistent with positive feedback as opposed to contrarian trading. ************************* TABLE X HERE ************************* The results are robust to using different benchmark returns to calculate excess returns and to using normalized order imbalances rather than buyers ratios to form the sell, hold, and buy portfolios. When we consider the alternative definition of speculative trading or sort stocks into four portfolios, the zero-cost portfolio’s return during the week after portfolio formation is still positive, but no longer statistically significant; the other results are little changed. To address concerns that illiquid stocks drive the results, we confine the analysis to stocks listed on the Neuer Markt, the segment for growth and technology stocks of the Frankfurt Stock Exchange. Neuer Markt stocks have to designate two exchange-approved sponsors that guarantee a liquid market – by maintaining an active limit order book and a certain maximum spread. Gerke and Bosch (1999) report that designated sponsors enhance liquidity, especially in less liquid Neuer Markt stocks and during periods of high market-wide spreads. The unreported returns of the speculative buy minus sell portfolio of Neuer Markt stocks are similar to those obtained for the full set of stocks. 29 These results are hard to square with the liquidity demand hypothesis. If the positive same-period correlation were due to market makers accommodating retail price pressure, one would expect to see prices reverting as the price pressure eases. While it is possible that the speculators exert persistent price pressure, it is unclear why serially correlated liquidity demands are not anticipated by professional liquidity suppliers. So far, our analysis has focused on speculative market orders. To check whether the distinction between speculative and non-speculative market orders matters, we repeat the portfolio exercise for non-speculative market orders. Panel A of Table XI shows the daily results corresponding to those in Table IX for speculative market orders. At the daily frequency, non-speculative buyers ratios are also positively correlated with yesterday’s, same-day, and tomorrow’s returns. However, the results are weaker than those for speculative market orders. For example, the return difference between the non-speculative buy and sell portfolios is 0.3% on the day of portfolio formation versus 1.8% for the speculative buy minus sell portfolio. At the weekly frequency, the differences are even more pronounced (see Panel B of Table XI). The zero-cost portfolio – long in stocks bought and short in stocks sold through non-speculative market orders – posts a return of -0.8% during the week prior to portfolio formation; recall that the corresponding zero-cost portfolio based on speculative orders returns 2.1%. In further contrast, stocks bought through non-speculative market orders fail to outperform stocks sold through such orders during the week following portfolio formation. ************************* TABLE XI HERE ************************* In sum, speculative retail imbalances are positively associated with contemporaneous and 30 future returns. These results cast doubt on hypotheses implying that the sample investors exert short-lived price pressure or are compensated for supplying liquidity to other market participants. The failure to detect longer-term reversals should be interpreted cautiously, however; due to the relatively short sample period, it is difficult to reliably estimate differences between buy and sell portfolio returns at horizons longer than one week. This makes it difficult to discriminate between the informed trading and noise trading hypotheses. If investors placing speculative bets were informed, one would expect that those who trade more speculatively – measured by the ratio of speculative trading volume to total trading volume – perform better than their peers. To check the plausibility of the informed trading hypothesis, one can compare the portfolio returns of the speculators with those of the other clients. After transaction costs, the quartile of the most aggressive speculators underperform their peers. This is primarily due to higher transaction costs as aggressive speculators also tend to be heavy traders. C Discussion At first glance, the results in the previous section appear to be at odds with the contrarian behavior reported for retail investors in Finnish stocks (Grinblatt and Keloharju (2000)), US stocks (Nofsinger and Sias (1999), Griffin, Harris, and Topaloglu (2003), Barber, Odean, and Zhu (2003), Kaniel, Saar, and Titman (2004)), the S&P 500 index (Goetzmann and Massa (2002)), and Australian stocks (Jackson (2003)). While differences in the data might account for some discrepancies, we suspect that the explanation lies in our focus on market orders in general, and speculative market orders in particular – all the papers referenced above combine market and limit orders. The advantage of focusing on speculative market orders is that they reflect the investors’ unadulterated 31 opinion about future prices in a timely manner. Limit orders may tell a different story. Especially unmonitored limit orders provide liquidity in response to large price movements in an almost automatic fashion. To illustrate, let us start from a given set of limit buy and sell orders on day zero. A large price decrease on day zero will make limit buy orders execute on day zero, implying a mechanical negative correlation between limit trading and contemporaneous returns. On day one, unless 1. traders with existing limit sell orders revise their price limit downwards or 2. new limit orders with lower price limits arrive or 3. the price increases by more than it decreased the previous day, only limit buy orders will be executed, and only if prices keep going down. Thus, unmonitored limit orders can explain the negative correlation between limit order imbalances and past returns. Without data on limit order submission, it is difficult to distinguish between active and mechanical contrarian behavior (see also Linnainmaa (2003)). ************************* TABLE XII HERE ************************* Thus, one would expect the relation between limit orders and returns to be substantially different from the relation between market orders and returns. Table XII (Panel A) confirms this conjecture. The same-day return of a zero-cost portfolio long in stocks bought through limit orders and short in stocks sold through limit orders during that day is negative and highly significant (−2.2%). There is no evidence for a price reversal on the subsequent day, however; returns of the zero-cost portfolio are slightly positive (0.06%), but not significantly different from zero. If we combine all trades executed through the brokerage (Panel B), the negative same-day effect of limit orders dominates - the mechanical negative correlation between limit orders and returns overwhelms the positive correlation between speculative 32 market orders and returns. ************************* TABLE XIII HERE ************************* The pattern difference between market and limit orders holds also at a weekly horizon (see Table XIII, Panel A). The portfolio of stocks most aggressively sold by limit traders has a contemporaneous return of 2.3% and the limit buy portfolio posts a return of -1.2%. As before, the zero-cost portfolio has a significant negative return of -3.5%. Interestingly, the limit buy portfolio outperforms during the week following portfolio formation whereas the limit sell portfolio underperforms; the return difference between the two portfolios – 0.4% – is statistically significant. These results are consistent with retail investors being compensated for providing liquidity to other market participants through limit orders, as suggested by Kaniel, Saar, and Titman (2004). Considering only limit orders, it is tempting to conclude that individuals follow a contrarian strategy at a daily and weekly frequency. The zero-cost portfolio based on limit orders has significantly negative returns in the period prior to portfolio formation (days: -0.9%, weeks: -1.7%; see Panel A of Tables XII and XIII). It seems likely that the classification of retail investors as contrarians in the prior literature is due to the investors’ use of limit orders. Without data on both submitted and executed limit orders, however, the negative correlation between imbalances and past returns is hard to interpret as it could be a reflection of “true” negative feedback trading or an artifact of unmonitored limit orders. The reported correlations between limit order imbalances and returns suggest that retail limit traders do indeed provide liquidity to other market participants. However, the strong 33 and possibly mechanical effect of limit orders masks the correlation between speculative trading and returns. Such heterogenous trading strategies might help explain conflicting evidence on imbalances and returns (see, e.g., Kaniel, Saar, and Titman (2004) vs Andrade, Chang, and Seasholes (2004) or Linnainmaa (2003)). IV Conclusion Retail investors act more similarly than would be expected by chance, even taking into account that retail investors will tend to move together because of short-sale constraints and unmonitored limit orders. One limitation of our study is that the observed investors are all clients at the same broker which might lead us to overstate the degree of comovement relative to a random sample of retail investors. However, since the correlation among the sample investors is not driven by explicit brokerage advice, investor location, IPOs underwritten by the sample broker, or automatic investment plans offered by the sample broker, there is good reason to believe that the observed trades are fairly representative of the population of self-directed German retail investors – the correlation between sample order imbalances and returns also point to this. Retail trades appear to be mainly coordinated by speculative motives; retail investors move together into stock A primarily because they share the belief that stock A’s price will appreciate more than that of other stocks, and not because stock A is part of a well-diversified savings portfolio. In this paper, we do not examine in detail what makes stock A a particularly attractive bet for retail investors - a difficult task that warrants future research. We do document that, based on market order imbalances, retail speculators as a group behave as positive feedback 34 traders, i.e., they buy recent winners and sell recent losers. We also document that limit order traders behave as if they were contrarians which could be because limit traders are contrarians or because many of their orders become stale and are mechanically executed. Future research with data on submitted, rather than just executed, limit orders could shed light on this interesting issue. Finally, correlated retail trading appears to affect price formation through two channels. First, order imbalances based on speculative market orders are positively correlated with contemporaneous as well as future returns – as a group, retail speculators appear to bet on the right direction. Second, limit order imbalances are negatively correlated with contemporaneous returns and positively correlated with future returns – limit order traders seem to be compensated for providing liquidity to other market participants. The speculators’ success should be put in perspective; given short-sale constraints, trading commissions, and price impact, mimicking the speculative buy minus sell portfolio may not result in actual profits. In fact, aggressive speculators do not appear to outperform their peers. At a minimum, however, the speculative trades taken together in our sample are not systematically on the wrong side of the market. 35 References Amihud, Yakov, and Haim Mendelson, 1980, Dealership market: market making with inventory, Journal of Financial Economics 8, 31–54. Andrade, Sandro, Charles Chang, and Mark Seasholes, 2004, Uninformed trading and asset prices, Working paper UC Berkely. Barber, Brad, and Terrance Odean, 2000, Trading is hazardous to your wealth: The common stock investment performance of individual investors, Journal of Finance 55, 773–806. , 2002, Online investors: Do the slow die first?, Review of Financial Studies 15, 455–487. Barber, Brad M., Terrance Odean, and Ning Zhu, 2003, Systematic noise, Working paper UCLA. Barberis, Nicholas C., and Andrei Shleifer, 2003, Style Investing, Journal of Financial Economics 68, 161–199. Campbell, John Y., Sanford J. Grossman, and Jiang Wang, 1993, Trading volume and serial correlation in stock returns, Quarterly Journal of Economics 108, 905–939. Chordia, Tarun, Richard Roll, and Avanidhar Subrahmanyam, 2005, Evidence on the speed of convergence to market efficiency, Journal of Financial Economics forthcoming. Chordia, Tarum, and Avanidhar Subrahmanyam, 2004, Order imbalance and individual stock returns, Journal of Financial Economics 72, 485–518. De Long, J. Bradford, Andrei Shleifer, Lawrence Summers, and Robert J. Waldmann, 1990a, Noise trader risk in financial markets, Journal of Political Economy 98, 703–738. 36 , 1990b, Positive feedback investment strategies and destabilizing rational speculation, Journal of Finance 45, 375–395. Deutsches Aktieninstitut, 2001, Factbook 2001, Frankfurt am Main. Feng, Lei, and Mark Seasholes, 2004, Correlated trading and location, Journal of Finance 59, 2117–2144. Gerke, Wolfgang, and Robert Bosch, 1999, Die Betreuer am Neuen Markt - eine empirische Analyse (Designated sponsors in the Neuer Markt - an empirical analysis, Working Paper, Center for Financial Studies, Frankfurt. Glosten, Lawrence R., and Paul R. Milgrom, 1985, Bid, ask and transaction prices in a specialist market with heterogeneously informed traders, Journal of Financial Economics 14, 71–100. Goetzmann, William N., and Masimo Massa, 2002, Daily momentum and contrarian behavior of index fund investors, Journal of Financial and Quantitative Analysis 32. Gorton, Gary, and George Pennacchi, 1993, Security baskets and index-linked securities, Journal of Business 66, 1–27. Griffin, John M., Jeffrey Harris, and Selim Topaloglu, 2003, The dynamics of institutional and individual trading, Journal of Finance 58, 2285–2320. Grinblatt, Mark, and Matti Keloharju, 2000, The investment behavior and performance of various investor types: a study of Finland’s unique data set, Journal of Financial Economics 55, 43–67. , 2001, What makes investors trade?, Journal of Finance 56, 589–616. 37 Grinblatt, Mark, Sheridan Titman, and Russ Wermers, 1995, Momentum investment strategies, portfolio performance, and herding: A study of mutual fund behavior, American Economic Review 85, 1088–1105. Grossman, S., and M. Miller, 1988, Liquidity and market structure, Journal of Finance 43, 617–633. Hong, Harrison, Terence Lim, and Jeremy C. Stein, 1999, Forecasting crashes: Trading volume, past returns and conditional skewness in stock prices, Journal of Financial Economics 54, 2143–2184. Huberman, Gur, and Werner Stanzl, 2004, Price manipulation and quasi-arbitrage, Econometrica 72, 1247–1275. Jackson, Andrew, 2003, The aggregate behaviour of individual investors, London Business School, Working Paper. Kaniel, Ron, Gideon Saar, and Sheridan Titman, 2004, Individual investor sentiment and stock returns, Working paper Faqua School of Business, Duke University. Kumar, Alok, and Charles Lee, 2005, Retail sentiment and return comovement, Journal of Finance forthcoming. Kyle, Albert S., 1985, Continuous auctions and insider trading, Econometrica 53, 1315–1335. Kyrolainen, Petri Juhani, and Jukka Perttunen, 2002, Investors’ activity and trading behavior, Working Paper, University of Oulu. Lakonishok, Josef, Andrei Shleifer, and Robert W. Vishny, 1992, The impact of institutional trading on stock prices, Journal of Financial Economics 32, 23–43. 38 Linnainmaa, Juhani, 2003, Who makes the limit order book? implications for contrarian strategies, attention-grabbing hypothesis, and the disposition effect., UCLA Working Paper. Nofsinger, John R., and Richard W. Sias, 1999, Herding and feedback trading by institutional and individual investors, Journal of Finance 54, 2263 – 2295. Odean, Terrance, 1998, Are investors reluctant to realize their losses?, Journal of Finance 53, 1775–1798. Shleifer, Andrei, and Robert Vishny, 1997, The limits to arbitrage, Journal of Finance 52, 35–55. Stoll, Hans R., 1978, The supply of dealership services in securities markets, Journal of Finance 33, 1133–1151. Subrahmanyam, Avanidhar, 1991, Risk aversion, market liquidity, and price efficiency, Review of Financial Studies 4, 417–441. Van Steenis, Huw, and Christian Ossig, 2000, European online investor, JP Morgan Equity Research Report. Wermers, Russ, 1999, Mutual fund herding and the impact on stock prices, Journal of Finance 54, 581–622. Wylie, Sam, 2005, Fund manager herding: A test of the accuracy of empirical results using uk data, Journal of Business 78. 39 A A Appendix Simulation of short-sale constraints First, all stocks are sorted each period into deciles by the number of clients who hold the stock at the beginning of the period; holder decile one contains the most widely held stocks and holder decile ten contains the least widely held stocks. To balance the number of transactions in each decile, exponentially more stocks are assigned to the deciles that contain less widely held stocks. Second, we simulate 1,000 data sets assuming that there is no correlated trading but there are short-sale constraints. For each stock j in holder decile Sj,t and period t, we randomly draw the number of initial buyers, active holders, and sellers, assuming that they are binomially distributed with #InitialBuyers(j, t) ∼ B (T otal#Investors(t) − #Holders(j, t), p̂newbuy (j, t)) #ActiveHolders(j, t) ∼ B (#Holders(j, t), p̂active (j, t)) #Sellers(j, t) ∼ B (#ActiveHolders(j, t), p̂sell (j, t)) where p̂newbuy (j, t) = #InitialBuyers(k, t) 1 X kSj,t k k∈S T otal#Investors(t) − #Holders(k, t) j,t 1 X #RepeatBuyers(k, t) + #Sellers(k, t) p̂active (j, t) = kSj,t k k∈S #Holders(k, t) j,t p̂sell (j, t) = 1 #Sellers(k, t) kSj,t k k∈S #RepeatBuyers(k, t) + #Sellers(k, t) X j,t 40 and #Holders(j, t) is the number of sample investors who hold stock j at the beginning of period t. Then, we calculate the average LSV measure across all stock-periods with at least two traders. The three probabilities vary considerably across holder deciles. For example, the ratio of actual initial buyers to potential initial buyers across all sample days is 0.013% for the most widely held stocks and 0.002% for the least widely held stocks; investors are more likely to buy into stocks that are already popular among their peers. Those who own less widely held stocks are more likely to trade them on any given day; the ratio of active holders to all holders is 3.5% for the least widely held stocks as opposed to 0.3% for the most widely held stocks. Moreover, owners of widely held stocks are more likely to increase their position than other owners; the ratio of owner-buyers to active owners in these stocks is 26.3% as opposed to 14.2% for the least widely held stocks.14 14 The probabilities and results are similar when the probabilities are estimated from “holder-weighted” as opposed to equally weighted relative frequencies, i.e., P k∈Sj,t #Sellers(k, t) 0 p̂sell (j, t) = P , k∈Sj,t (#RepeatBuyers(k, t) + #Sellers(k, t)) for example. 41 42 Unit [DEM ’000s] [%] [%] [%] [%] [%] [%] Number of portfolios 33483 32485 4142 1650 5300 32485 17604 Panel B: Trading statistics for the period Feb 1, 1998 - May 31, 2000 Trading volume [DEM ’000s] 37733 Fraction of individual stock trading [%] Fraction of stock fund trading [%] Fraction of bond trading [%] Fraction of money market trading [%] Fraction of options trading [%] Portfolio value Stocks as a fraction of total portfolio Bonds as a fraction of total portfolio Money market as a fraction of total portfolio Options as a fraction of total portfolio HHI of equity portfolio Stock funds as a fraction of all stocks Panel A: Portfolio statistics as of May 31, 2000 484 61.8% 21.3% 3.1% 3.3% 9.1% Mean 136 89.7% 3.4% 1.7% 4.2% 25.1% 29.5% 111 72.7% 2.8% 0% 0% 0% Median 45 100% 0% 0% 0% 13.5% 7.4% 18,262,772 63.4% 11.6% 2.3% 6.3% 16.4% Aggregate 4,557,253 87.6% 5.1% 3.9% 2.2% 0.0% 24.0% Panel A summarizes the sample portfolio statistics at the end of the sample period, May 31, 2000. The “Number of portfolios” is the number of accounts for which the statistic is non-zero, but the mean and median are calculated across all investors. For example, there are 4,142 clients who hold bonds, and the average bond holding across all 33,483 clients with non-zero portfolio holding as of May 31, 2000 is 3.4%. Panel B summarizes the sample trading statistics for the period February 1, 1998 until May 31, 2000. There are more accounts with trading activity during the sample period than there are accounts with holdings because 1. some clients close their accounts and 2. some active clients may just hold cash in their accounts. During the sample period, the exchange rate between Deutsche Mark [DEM] and US Dollar [USD] is approximately 2 DEM per USD. “Aggregate” refers to entire brokerage portfolio. “Stocks” refers to all domestic and foreign stocks as well as all equity mutual funds (same for “Bonds”). The HHI is defined as the sum of squared portfolio weights. A portfolio consisting of n 2 stocks equally weighted would have an HHI of n · n1 . Note that we assume equity mutual funds to consist of one hundred equally weighted positions that do not overlap with other holdings of the investor. That is, the HHI of portfolio of an investor holding one mutual fund is 1% and that of an investor splitting his money equally between two mutual funds is 0.5%. “Trading volume” is the sum of the absolute value of purchases and sales. Table I: Summary statistics of sample investor portfolios and trading 43 4.8% 4.3% 17.6% 47,341 5.4% 4.3% 16.2% 21,176 6.4% 5.0% 15.3% 7,552 8.3% 7.0% 15.4% 3,288 Panel A: Daily LSV - mean LSV - median LSV - std Number of observations Panel B: Weekly LSV - mean LSV - median LSV - std Number of observations Panel C: Monthly LSV - mean LSV - median LSV - std Number of observations Panel D: Quarterly LSV - mean LSV - median LSV - std Number of observations (1) All observations 7.5% 6.6% 15.8% 1,992 5.5% 3.8% 15.5% 5,348 4.6% 3.6% 16.9% 15,585 (2) Bottom trading quartiles 4.0% 3.5% 18.5% 35,209 9.4% 7.9% 13.5% 870 8.7% 6.9% 13.5% 1,995 7.5% 5.6% 13.7% 5,429 (3) Top trading quartile 7.1% 5.3% 14.6% 11,950 8.1% 7.8% 9.2% 639 6.2% 6.0% 7.5% 631 6.0% 5.2% 7.8% 616 6.1% 4.8% 8.5% 591 (4) Cross section 6.9% 6.3% 4.0% 88 6.1% 5.5% 3.6% 87 4.8% 4.8% 2.7% 87 4.5% 3.9% 4.8% 88 (5) Industry 10.3% 8.2% 17.1% 3,288 9.2% 6.7% 17.6% 7,552 7.6% 5.4% 18.0% 21,176 (6) Alternative control for aggregate liquidity effects 6.5% 4.8% 19.0% 47,341 This table summarizes daily, weekly, monthly, and quarterly LSV measures (see Lakonishok, Shleifer, and Vishny (1992)) of correlated trading by the sample investors in German stocks, as described in the text. The sample period is February 1, 1998 to May 31, 2000. To form the top trading quartile (“top quartile”), we take each stock and identify the quarter of trading periods with the highest numbers of active traders in the stock. The bottom three trading quartiles (“bottom quartiles”) contain the remaining observations. Cross-sectional LSV measures are calculated by averaging LSV measures first across the observations for a given stock, then across stocks. To calculate industry LSV measures, we assign to each stock the Level 6 Datastream industry classification (Level 6 is the most detailed classification). LSV measures are then calculated for each industry rather than for each stock. In Column (6), we use the period-average buyers ratio across all equity securities – not just the sample stocks – as a control for aggregate liquidity effects. All estimates are significantly different from zero at conventional significance levels assuming that the observations are independent. Table II: The LSV measure of correlated trading: baseline results Table III: Correlated trading due to short-sale constraints This table summarizes the bias of the LSV measure due to short-sale constraints (see Wylie (2005)). The original data set is resampled 1,000 times assuming that the sample investors trade randomly, but face short-sale constraints. Quartiles of observations based on trading activity are formed as in Table II. The mean and standard deviation of the LSV measures are calculated across the 1,000 simulations. All estimates are significantly different from zero at conventional significance levels. Panel A: Daily LSV bias - mean LSV bias - std All observations 2.31% 0.06% Bottom quartiles 2.16% 0.08% Top quartile 2.75% 0.08% Panel B: Weekly LSV bias - mean LSV bias - std 1.89% 0.08% 1.75% 0.10% 2.28% 0.11% Panel C: Monthly LSV bias - mean LSV bias - std 1.75% 0.10% 1.65% 0.13% 2.01% 0.15% Panel D: Quarterly LSV bias - mean LSV bias - std 1.79% 0.16% 1.60% 0.20% 2.27% 0.19% 44 45 All 4.4% 3.9% 17.8% 26,605 5.1% 4.4% 16.4% 13,495 5.9% 4.8% 15.3% 5,230 7.6% 6.9% 16.0% 2,328 Panel A: Daily LSV - mean LSV - median LSV - std Number of observations Panel B: Weekly LSV - mean LSV - median LSV - std Number of observations Panel C: Monthly LSV - mean LSV - median LSV - std Number of observations Panel D: Quarterly LSV - mean LSV - median LSV - std Number of observations 7.6% 6.9% 16.2% 1,302 5.3% 3.7% 15.8% 3,646 4.4% 3.6% 17.2% 9,840 7.5% 5.7% 13.3% 575 7.3% 6.3% 12.9% 1,375 6.9% 5.2% 13.4% 3,455 Market orders Bottom trading quartiles Top trading quartile 3.7% 6.2% 2.9% 4.8% 18.8% 14.5% 19,713 6,722 7.5% 6.1% 15.6% 2,955 6.4% 5.0% 16.0% 6,556 5.9% 5.1% 17.3% 16,556 All 5.8% 6.2% 18.4% 29,331 6.8% 5.8% 15.7% 1,769 5.6% 4.2% 16.4% 4,616 5.1% 4.3% 17.9% 12,130 Limit orders Bottom trading quartiles 4.9% 5.8% 19.2% 21,735 9.0% 7.0% 13.9% 767 8.6% 6.7% 14.1% 1,737 8.2% 6.5% 14.9% 4,258 Top trading quartile 8.3% 7.0% 15.8% 7,430 This table summarizes daily, weekly, monthly, and quarterly LSV measures of correlated trading, calculated separately for market orders and for limit orders. The sample period is February 1, 1998 to May 31, 2000. Quartiles of observations based on trading activity are formed as in Table II. The LSV biases due to short-sale constraints, estimated by re-sampling the original data set 1,000 times (and assuming that the sample investors trade randomly, but face short-sale constraints), are very similar to those reported in Table III and thus omitted. All estimates are significantly different from zero at conventional significance levels assuming that the observations are independent across stocks and time. Table IV: Correlated trading due to limit orders Table V: Summary statistics of speculative and non-speculative orders This table summarizes statistics of individual German stock trades classified as speculative and non-speculative according to the baseline definition and the alternative definition of speculative trading (as described in Section C.1). The sample period is February 1, 1998 to May 31, 2000. Overall Fraction of market orders Purchases Number DEM volume 299,991 2,649,264,636 52.3% 47.5% Number 224,761 51.6% Sales DEM volume 2,638,519,351 48.2% Fraction of trading classified as Speculative - baseline definition Speculative - alternative definition 57.6% 52.2% 68.3% 61.2% 63.8% 50.7% 62.3% 51.4% Non-speculative - alternative definition thereof: Savings/Dissavings driven Liquidity driven Diversification driven Allocation driven Rebalancing driven Hedging driven Ambiguous 72.7% 2.9% n/a 0.9% 23.3% n/a 0.1% 62.2% 5.1% n/a 1.2% 31.5% n/a 0.1% 27.4% 1.8% 9.6% 0.3% 38.4% 0.8% 21.8% 25.2% 3.6% 6.7% 0.3% 46.0% 0.5% 17.7% 46 Table VI: Correlation of speculative and non-speculative trades This table summarizes daily, weekly, monthly, and quarterly LSV measures of correlated trading, calculated separately for speculative market orders and for non-speculative market orders. Orders are classified as speculative or non-speculative according to the baseline definition and the alternative definition of speculative trading described in detail in Section C.1. The sample period is February 1, 1998 to May 31, 2000. The LSV biases due to short-sale constraints, estimated by re-sampling the original data set 1,000 times (and assuming that the sample investors trade randomly, but face short-sale constraints), are very similar to those reported in Table III and are thus omitted. All estimates are significantly different from zero at conventional significance levels. Panel A: Daily LSV - mean LSV - median LSV - std Number of observations Baseline definition of speculative trading (1) (2) Non-speculative Speculative market orders market orders 4.7% 4.8% 5.8% 5.0% 18.3% 18.5% 10,378 19,094 Alternative definition of speculative trading (3) (4) Non-speculative Speculative market orders market orders 4.3% 4.8% 5.3% 5.3% 18.3% 18.8% 11,750 16,647 Panel B: Weekly LSV - mean LSV - median LSV - std Number of observations 4.9% 5.7% 17.7% 7,629 6.1% 6.0% 17.3% 10,703 4.5% 5.1% 17.4% 8,198 5.6% 5.3% 17.6% 9,829 Panel C: Monthly LSV - mean LSV - median LSV - std Number of observations 4.8% 4.5% 16.0% 3,842 8.2% 7.6% 16.6% 4,463 4.5% 4.1% 15.7% 3,974 7.3% 6.3% 16.4% 4,163 Panel D: Quarterly LSV - mean LSV - median LSV - std Number of observations 5.5% 4.8% 15.4% 1,930 10.1% 9.3% 16.6% 2,117 5.6% 5.0% 15.3% 1,976 9.2% 8.3% 16.3% 2,009 47 48 Hypotheses Return chasing Contrarian trading Informed trading Liquidity demand Liquidity supply Noise trading Implication of the hypothesis for the correlation between observed buyers ratios and Past returns Contemporaneous returns Short-term future returns Longer-term future returns + 0 0 0 0 0 0 ? + + 0 ? + 0 ? + 0 ? + + or 0 - The table shows the sign of the correlation between trading direction and past, contemporaneous, and future returns predicted by the hypotheses discussed in section 4.1. Table VII: Hypotheses about the relation between buyers ratios and returns Table VIII: Average buyers ratios for sell, hold, and buy portfolios Each period, we sort the sample stocks by the observed buyers ratio based on speculative market orders. According to its buyers ratio rank, each stock is sorted into one of three portfolios of roughly equal size, i.e., similar number of stocks: the sell portfolio (PF1), the hold portfolio (PF2), and the buy portfolio (PF3). The first line corresponding to a portfolio shows the average buyers ratio of the portfolio across the sample period 2/1/98 - 5/31/00; within a portfolio, stocks are equally weighted. The second lines report standard deviations. Column (0) shows average buyers ratios for the portfolio formation period; in addition, column (0) also shows the summary statistics of the portfolio cutoffs (reported under the heading “lower cutoff” and “upper cutoff”). Column (-1) shows the average buyers ratios of the sell, hold, and buy portfolios during period preceding portfolio formation; Column (+1) shows the average buyers ratios of the sell, hold, and buy portfolios during the period following portfolio formation. Panel A shows daily results and Panel B shows weekly results. Panel A Horizon (days) 0 (formation) -1 lower cutoff PF1 PF2 PF3 PF3-PF1 0.416 (0.21) 0.499 (0.18) 0.580 (0.19) 0.164 (0.26) 0 0.293 (0.19) 0.607 (0.16) Panel B 0.130 (0.15) 0.459 (0.15) 0.812 (0.14) 0.683 (0.19) lower cutoff PF1 PF2 PF3 PF3-PF1 0.405 (0.10) 0.458 (0.09) 0.521 (0.10) 0.116 (0.13) 0.293 (0.19) 0.607 (0.16) 1 - Horizon (weeks) 0 (formation) -1 0 0.282 (0.12) 0.557 (0.08) 0.396 (0.20) 0.488 (0.19) 0.555 (0.19) 0.159 (0.26) +1 upper cutoff 0.100 (0.08) 0.433 (0.08) 0.772 (0.08) 0.672 (0.10) 49 +1 upper cutoff 0.282 (0.12) 0.557 (0.08) 1 - 0.373 (0.11) 0.438 (0.08) 0.496 (0.08) 0.123 (0.13) Table IX: Portfolios based on daily comovement: speculative trades At the end of each trading day, we sort all stocks which are bought or sold net within the brokerage according to the baseline definition of speculative trading by their buyers ratio. We then form three portfolios (PF1 to PF3) and calculate the equally weighted excess return – over a holding-weighted index of all German stocks that represent more than 0.5% of the brokerage’s combined German stock portfolio – of these portfolios on the day of portfolio formation (Column 0). Column (+1) reports the excess return of the same portfolio during the next trading day, Column (-1) during the previous trading day. Lines 1-3 show the excess returns (in percent) of the three portfolios, line 4 shows the return of a zero-cost portfolio P3-P1, long in stocks bought within the brokerage and short in stocks sold. Newey-West standard errors (two lags) are reported beneath average returns. Horizon (days) 0 1 -1 PF1 (Brokerage sell) PF2 PF3 (Brokerage buy) PF3-PF1 N 0.047 (0.08) 0.789 (0.18) 1.640 (0.18) -0.222 (0.07) 0.368 (0.09) 1.588 (0.14) -0.119 (0.06) 0.157 (0.08) 0.445 (0.10) 1.593 (0.19) 1.810 (0.16) 0.564 (0.11) 609 609 609 Table X: Portfolios based on weekly comovement: speculative trades At the end of each week, we sort all stocks which are bought or sold net within the brokerage according to the baseline definition of speculative trading by their buyers ratio. We then form three portfolios (PF1 to PF3) and calculate the equally weighted excess return – over a holding-weighted index of all German stocks that represent more than 0.5% of the brokerage’s combined German stock portfolio – of these portfolios during the week of portfolio formation (Column 0). Column (+1) reports the excess return of the same portfolio during the next week, Column (-1) during the previous week. Lines 1-3 show the excess returns (in percent) of the three portfolios, line 4 shows the return of a zero-cost portfolio P3-P1, long in stocks bought within the brokerage and short in stocks sold. Newey-West standard errors (two lags) are reported beneath average returns. -1 PF1 (Brokerage sell) PF2 PF3 (Brokerage buy) PF3-PF1 N Horizon (weeks) 0 1 0.155 (0.22) 0.834 (0.28) 2.236 (0.49) -1.162 (0.21) 0.188 (0.26) 3.516 (0.45) 0.023 (0.26) 0.021 (0.23) 0.636 (0.27) 2.081 (0.50) 4.678 (0.49) 0.612 (0.30) 122 122 122 50 Table XI: Portfolios based on daily and weekly comovement: non-speculative trades At the end of each period, we sort all stocks which are bought or sold through non-speculative market orders, according to the baseline definition of speculative/non-speculative trading, by their buyers ratio. We then form three portfolios (PF1 to PF3) and calculate the equally weighted excess return – over a holding-weighted index of all German stocks that represent more than 0.5% of the brokerage’s combined German stock portfolio – of these portfolios during the period of portfolio formation (Column 0). Column (+1) reports the excess return of the same portfolio during the following period, Column (-1) during the previous period. Lines 1-3 show the excess returns (in percent) of the three portfolios, line 4 shows the return of a zero-cost portfolio P3-P1, long in stocks bought within the brokerage and short in stocks sold. Panel A reports the results for the daily frequency; Panel B reports the corresponding results for the weekly frequency. Newey-West standard errors (two lags) are reported beneath average returns. Panel A: Daily Days relative to portfolio formation -1 0 1 PF1 (Brokerage sell) 0.737 (0.15) 1.007 (0.17) 1.315 (0.22) 0.358 (0.08) 0.487 (0.10) 0.657 (0.09) -0.185 (0.07) 0.019 (0.06) 0.245 (0.08) 0.579 (0.25) 0.298 (0.12) 0.430 (0.11) 609 609 609 PF2 PF3 (Brokerage buy) PF3-PF1 N Panel B: Weekly Weeks relative to portfolio formation -1 0 1 PF1 (Brokerage sell) 1.758 (0.41) 1.112 (0.29) 0.966 (0.26) 0.626 (0.26) 1.209 (0.29) 1.374 (0.29) 0.212 (0.23) 0.037 (0.20) 0.171 (0.19) -0.792 (0.46) 0.748 (0.25) -0.041 (0.28) 122 122 122 PF2 PF3 (Brokerage buy) PF3-PF1 N 51 Table XII: Portfolios based on daily comovement: limit orders and all orders At the end of each trading day, we sort all stocks which are bought or sold net within the brokerage by their buyers ratio. We then form three portfolios (PF1 to PF3) and calculate the equally weighted excess return – over a holding-weighted index of all German stocks that represent more than 0.5% of the brokerage’s combined German stock portfolio – of these portfolios on the day of portfolio formation (Column 0). Column (+1) reports the excess return of the same portfolio during the next trading day, Column (-1) during the previous trading day. Lines 1-3 show the excess returns (in percent) of the three portfolios, line 4 shows the return of a zero-cost portfolio P3-P1, long in stocks bought within the brokerage and short in stocks sold. The statistics in Panel A are based on buyers ratios using limit orders only; the statistics in Panel B are based on buyers ratios using all orders. Newey-West standard errors (two lags) are reported beneath average returns. Panel A: Limit orders only -1 PF1 (Brokerage sell) PF2 PF3 (Brokerage buy) PF3-PF1 N 0.845 (0.07) 0.481 (0.08) -0.020 (0.07) 1.541 (0.08) 0.309 (0.08) -0.629 (0.07) 0.119 (0.06) 0.166 (0.06) 0.177 (0.06) -0.865 (0.08) -2.170 (0.09) 0.058 (0.07) 609 609 609 Panel B: All orders -1 PF1 (Brokerage sell) PF2 PF3 (Brokerage buy) PF3-PF1 N 52 Horizon (days) 0 1 Horizon (days) 0 1 0.319 (0.05) 0.355 (0.06) 0.239 (0.06) 0.636 (0.05) 0.281 (0.06) -0.003 (0.06) -0.012 (0.05) 0.071 (0.05) 0.295 (0.05) -0.080 (0.06) -0.639 (0.07) 0.307 (0.06) 609 609 609 Table XIII: Portfolios based on weekly comovement: limit orders and all trades At the end of each week, we sort all stocks which are bought or sold net within the brokerage by their buyers ratio. We then form three portfolios (PF1 to PF3) and calculate the equally weighted excess return – over a holding-weighted index of all German stocks that represent more than 0.5% of the brokerage’s combined German stock portfolio – of these portfolios during the week of portfolio formation (Column 0). Column (+1) reports the excess return of the same portfolio during the week following portfolio formation, Column (-1) during the week preceding portfolio formation. Lines 1-3 show the excess returns (in percent) of the three portfolios, line 4 shows the return of a zero-cost portfolio P3-P1, long in stocks bought within the brokerage and short in stocks sold. The statistics in Panel A are based on buyers ratios using limit orders only; the statistics in Panel B are based on buyers ratios using all orders. Newey-West standard errors (two lags) are reported beneath average returns. Panel A: Limit orders only -1 PF1 (Brokerage sell) PF2 PF3 (Brokerage buy) PF3-PF1 N 1.310 (0.21) 0.852 (0.22) -0.409 (0.24) 2.290 (0.25) 0.669 (0.26) -1.177 (0.23) -0.154 (0.22) 0.334 (0.25) 0.269 (0.21) -1.719 (0.19) -3.467 (0.24) 0.424 (0.20) 122 122 122 Panel B: All orders -1 PF1 (Brokerage sell) PF2 PF3 (Brokerage buy) PF3-PF1 N 53 Horizon (weeks) 0 1 Horizon (weeks) 0 1 0.625 (0.22) 0.402 (0.20) 0.136 (0.23) 0.714 (0.23) 0.408 (0.20) 0.198 (0.25) -0.199 (0.22) -0.027 (0.20) 0.409 (0.21) -0.489 (0.22) -0.516 (0.25) 0.608 (0.17) 122 122 122 Previous DNB Working Papers in 2005 No. 27 No. 28 No. 29 No. 30 No. 31 No. 32 No. 33 No. 34 No. 35 No. 36 No. 37 No. 38 No. 39 No. 40 No. 41 No. 42 No. 43 No. 44 No. 45 No. 46 No. 47 No. 48 No. 49 No. 50 No. 51 No. 52 No. 53 No. 54 No. 55 No. 56 No. 57 No. 58 No. 59 No. 60 No. 61 No. 62 Jan Marc Berk and Beata K. Bierut, On the optimality of decisions made by hup-and-spokes monetary policy committees Ard H.J. den Reijer, Forecasting dutch GDP using large scale factor models Olivier Pierrard, Capital Market Frictions, Business Cycle and Monetary Transmission Jan Willem van den End and Mostafa Tabbae, Measuring financial stability; applying the MfRisk model to the Netherlands Carin van der Cruijsen and Maria Demertzis, The Impact of Central Bank Transparency on Inflation Expectations Allard Bruinshoofd, Bertrand Candelon and Katharina Raabe, Banking Sector Strength and the Transmission of Currency Crises David-Jan Jansen and Jakob de Haan, Were verbal efforts to support the euro effective? A highfrequency analysis of ECB statements Riemer P. Faber and Ad C.J. Stokman, Price convergence in Europe from a macro perspective: Product categories and reliability Jan Kakes and Cees Ullersma, Financial acceleration of booms and busts Wilko Bolt and David B. Humphrey, Public Good Aspects of TARGET: Natural Monopoly, Scale Economies, and Cost Allocation Piet van Gennip, Loan extension in China: a rational affair? Joël van der Weele, Financing development: debt versus equity Michael Hurd and Susann Rohwedder, Changes in Consumption and Activities at Retirement Monica Paiella and Andrea Tiseno, Stock market optimism and participation cost: a mean-variance estimation Monika Bütler, Olivia Huguenin and Federica Teppa, What Triggers Early Retirement? Results from Swiss Pension Funds Günther Fink and Silvia Redaelli, Understanding Bequest Motives – An Empirical Analysis of Intergenerational Transfers Charles Grant and Tuomas Peltonen, Housing and Equity Wealth Effects of Italian Households Philipp Maier, A ‘Global Village’ without borders? International price differentials at eBay Robert-Paul Berben and Teunis Brosens, The Impact of Government Debt on Private Consumption in OECD Countries Daniel Ottens and Edwin Lambregts, Credit Booms in Emerging Market Economies: A Recipe for Banking Crises? Jaap Bikker and Michiel van Leuvensteijn, An exploration into competition and efficiency in the Dutch life insurance industry Paul Cavelaars, Globalisation and Monetary Policy Janko Gorter, Subjective Expectations and New Keynesian Phillips Curves in Europe Ralph de Haas and Ilko Naaborg, Does Foreign Bank Entry Reduce Small Firms’ Access to Credit? Evidence from European Transition Economies Ralph de Haas and Ilko Naaborg, Internal Capital Markets in Multinational Banks: Implications for European Transition Countries Roel Beetsma, Massimo Guiliodori and Franc Klaassen, Trade Spillovers of Fiscal Policy in the European Union: A Panel Analysis Nicole Jonker, Payment Instruments as Perceived by Consumers – A Public Survey Luis Berggrun, Currency Hedging for a Dutch Investor: The Case of Pension Funds and Insurers Siem Jan Koopman, André Lucas and Robert J. Daniels, A Non-Gaussian Panel Time Series Model for Estimating and Decomposing Default Risk Guay C. Lim and Paul D. McNelis, Real Exchange Rate and Current Account Dynamics with Sticky Prices and Distortionary Taxes Roel Beetsma, Alex Cukierman and Massimo Giuliodori, Wars, Redistribution and Civilian Federal Expenditures in the U.S. over the Twentieth Century S. Fabiani, M. Druant, I. Hernando, C. Kwapil, B. Landau, C. Loupias, F.Martins, T. Mathä, R. Sabbatini, H. Stahl and A. Stokman, The Pricing Behaviour of Firms in the Euro Area: New Survey Evidence Jan Marc Berk and Beata K. Beirut, Communication in Monetary Policy Committees Robert-Paul Berben and W. Jos Jansen, Bond Market and Stock Market Integration in Europe Leo de Haan and Elmer Sterken, Asymmetric Price Adjustment in the Dutch Mortgage Market L. J. Álvarez , E. Dhyne, M. Hoeberichts, C. Kwapil, H. Le Bihan, P. Lünnemann, F. Martins, R. Sabbatini, H. Stahl, P. Vermeulen and J. Vilmunen, Sticky prices in the euro area: a summary of new micro evidence Previous DNB Working Papers in 2005 (continued) No. 63 No. 64 No. 65 No. 66 No. 67 No. 68 No. 69 No. 70 No. 71 Peter Vlaar, Defined benefit pension plans and regulation Hans Fehr and Christian Habermann, Risk sharing and efficiency implications of progressive pension arrangements Axel Börsch-Supan, Alexander Ludwig and Joachim Winter, Aging, Pension Reform, and Capital Flows: A Multi-Country Simulation Model Zvi Bodie, Pension Insurance Frank de Jong, Valuation of pension liabilities in incomplete markets Vincent Bodart, Olivier Pierrard and Henri R. Sneessens, Calvo Wages in a Search Unemployment Model Jacob A. Bikker, Laura Spierdijk and Pieter Jelle van der Sluis, Cheap versus Expensive Trades: Assessing the Determinants of Market Impact Costs Wilko Bolt and Alexander F. Tieman, On myopic equilibria in dynamic games with endogenous discouting Wilko Bolt, David Humphrey and Roland Uittenbogaard, The Effect of Transaction Pricing on the Adoption of Electronic Payments: A Cross-Country Comparison