Investments and Financial Markets

Transcription

1 Investments and Financial Markets Part III: Market Efficiency UE 106 Master SOM 1 Market Efficiency Chapter Informational efficiency (and costs) 2. Price forecast and past prices 3. Production of financial information 4. Information, price movements and event-study methodology 5. Adjustment speed 6. Market anomalies Exercises 2 Université Paris-Dauphine 1

2 Market Efficiency 1. Informational efficiency (and costs) Forms of efficiency Is the market for price forecasts competitive? Costs associated with price forecasting Can equilibrium prices reveal information fully? Delayed reaction in an informationally efficient market Portfolio management and informational efficiency Rational expectations and efficiency Chapter 20 3 Forms of efficiency Efficient frontier Locus of portfolios that maximize expected returns subject to volatility constraint (Markowitz, part II) Allocational efficiency Is the market pareto efficient? Organizational efficiency Optimal market model Operational efficiency Transaction costs Informational efficiency 4 Université Paris-Dauphine 2

3 Informational efficiency The following two questions are equivalent: 1) Are price forecasts profitable? «In an efficient market, prices reflect information to the point where the marginal benefits of acting on information (the profits to be made) do not exceed the marginal costs» - Fama (1991, p. 1576) 2) Is the market for price forecasts competitive? Profits earned from forecasting activity simply cover the costs paid by analysts The social role of analysts producing price forecasts: Facilitate the incorporation of information into prices and produce (although not intentionally) a public good The revenues they earn thus make sense (from a social viewpoint) The market is (informationally) efficient if: Price forecasts are not profitable (no profit does not imply no salary!) The competition between professionals leads to zero-profits Forecasts by analysts make stock prices valuable 5 Forecast costs Producing and exploiting information is costly: Salaries The cost of acquiring information Transaction costs (higher for actively-managed portfolios) Even if price forecasts result in profits from the trade: The question is: Are the profits from the trade large enough to cover information processing costs? If markets are efficient, the profits from trade just cover costs (no profit left) 6 Université Paris-Dauphine 3

4 Do equilibrium prices reveal information fully? (I) Trading prices cannot fully incorporate all value-relevant information (impossibility of fully-revealing equilibrium) Why? Frictions and costs Informed investors are better-off if they are not detected by other traders. They try to camouflage and avoid using large visible orders (Kyle, 1985) Implications? Departures from true prices may exist since arbitraging these "errors" would lead to zero profit once costs are accounted for «We have argued that because information is costly, prices cannot perfectly reflect the information which is available, since if it did, those who spent resources to obtain it would receive no compensation» Grossman S. et J. Stiglitz, Revealing equilibrium? (II) In the presence of frictions (costs and delays) 1. The "true" value is unknown 2. Only prices are observed 3. Prices evolve as if the "true" value were bracketed by invisible barriers 4. Arbitrageurs trade when prices cross the barriers Buy «assume that every professional has the same expectations, and the same opportunity costs. Then prices will behave as a random walk with reflecting barriers». Cootner, 1964, p Université Paris-Dauphine 4

5 Speed of reaction in an efficient market «instantaneous and full adjustment» is IMPOSSIBLE in an efficient market (with adjustment costs) price Are the two following examples consistent with market efficiency? Which costs? Which frictions? Info time price cours price cours Info temps time Info temps time 9 Portfolio management and efficiency In the presence of costs, price forecasts are possible (and salaries can be paid) Profits vanish due to competition Social role of analysts (those who make prices valuable, public good) The development of index funds ETFs (most of which are passive) Highlights how "beating the market" is difficult Portfolio management professionals acknowledge that markets are efficient 10 Université Paris-Dauphine 5

6 Forms of expectations and market efficiency Naïve expectations X a + = t, t 1 t a Rational expectations (Muth, 1961) E[ X I ] Forecast error (made at t and observed at t+1) X The expectation of the forecast error is 0 E[Error ] = E[X E[X I ]] = 0 t t+ 1 t+ 1 t I t I t + 1 X t,t+ 1 = t + 1 Error = X X a t t+ 1 t,t+ 1 Law of iterated expectations ( ) E[ x] E E x y = Exercise: assuming people's expectations are rational «The best estimate of tomorrow's price is todays's price». True or False? t 11 Price forecasts and past prices Random walk The methods of technical analysis Principles of technical analysis Moving averages Simulation of a strategy Technical analysis: Conclusion? 12 Université Paris-Dauphine 6

7 Generating random numbers Then select cells A8:C8 and copy them towards the bottom 13 Pernod Ricard a Random walk? b c Which is the stock? a, b or c? The starting value is equal to 330, the expected return as well as the volatility are those observed for the stock over the period Université Paris-Dauphine 7

8 Signal or random walk? eye tusk ear trunk Shoulder Head Shoulder 94 Neck 93 Sell signal Signal? J. Hamon 15 Elliott wave approach Weekly data Rising band The structure in three waves, characterizes a consolidation period, defined by the increase from 1935 to 2610 points, leads us to think that there is downward directional pressure» «We consider 2325/2330 as the support level, which corresponds to 2nd Fibonacci retracement, i.e. the rally from 2050 to 2605/2610 observed in July 2012.» H. Naka, Agefi, 19 Nov Université Paris-Dauphine 8

9 10 December 2012 (Agefi) Weekly data The recent evolution strengthens our analysis that the bearish scenairo has been defended by the support levels since last June. Potential rise up to 3890/3915 points starting within the month. 17 Methods of technical analysis Information used Historical data prices, volumes, open interest Underlying principles The question is: when should I buy/sell? The market reflects (with some delay) private information History tends to repeat itself Variety of approaches Evolve over time with communication and information processing technologies Not necessarily based on graphs (indicator) From 'tape-reading trader' in Lefèvre (1923) to algorithmic trading or pairs trading. See Béchu, Bertrand and Nebenzhal, 2008, Economica 18 Université Paris-Dauphine 9

10 Steps Step 1: select a type of chart Bar chart Point and figure Candelstick Market profile, Kagi, Renko, Linebreak charts, etc. Step 2: identify signals Identifying characteristic patterns Filtering See also downloadable document «Graphique 2013.nb» 19 Identifying signals (buy/sell) identifying reversals in trends Characteristic patterns Head and shoulders in a bar chart 'Congestion zones' in a point and figures chart, Tombstones in candelstick charts Fibonacci's retracings for Elliot's waves Filtering examples: Alexander's filter (% applied to latest highest / lowest) Moving averages Fourier transforms (used by some traders since 40 years!) 20 Université Paris-Dauphine 10

11 Moving averages on Lafarge stock Potential signals: crossing points / changes in slope 21 Lafarge (cont'd) Many crossing points imply high transaction costs Applying a ± 2% confidence interval around the 50-day moving average reduces the number of bad signals 22 Université Paris-Dauphine 11

12 Simulating a strategy (backtesting) Tests must be performed on individual stocks (not indices) It is necessary to account for trading costs and frictions: Buys are completed at the ask and sales at the bid. Technical methods generate more signals when volatility is high and bid-ask spread is large Account for delays in trades Necessary to account for dividends (dividends are cashed-in by the buy-and-hold investment strategy) Beware of data-mining On a given time period it is always possible to find a method that yields very outstanding predictions Perform out-of-sample tests Chapter 20, section Technical analysis: conclusion? It is impossible to prove that technical analysis is completely misleading All we can do is make experiments Based on published recommendations By simulating the methods using historical data Conclusions reported in academic journals are always negative (see chapter 20) However: suppose you find a secret formula that allows you to earn millions of dollars. Would you publish your results in an academic journal? Maybe academics are only aware of useless techniques! 24 Université Paris-Dauphine 12

13 Production of financial information Information Sensitive information and insider trading Public information Private information Financial analysts Types of analysts Analyst recommendations and consensus The added value of analysts 25 Sensitive information and insider trading A sensitive information is information that is acquired in the context of professional activities and is such that its publication would cause price movements Trading on sensitive information is prohibited Information on the accounts that will be made public in a near future Information on merger/acquisition deals between two firms Information on order flow (trader X buys) Ex: Frontpoint (hedge fund), specialized in funds investing in health industry A doctor (web of experts) informs the fund that clinical trials on the Albuferon medicine are a failure Arrested on Nov. 2010, released on bail for $3 million 26 Université Paris-Dauphine 13

14 Insider trading, price manipulation Hanif Lalani and BT s Net2S takeover bid, Oct Persons related with insider trading were fined a total of 6.17 million by AMF. Three hedge funds based on Virgin Islands (CIF, Coudrée and CMA) were fined for 6.2m on May 2012 for short-selling and not giving back Natixis shares within the agreed time period. SGAM (Societe Generale Asset Management) was fined 1m in Oct 2011 for failing to comply with internal risk regulations. Carmignac Asset Management fined 500,000 for insufficient information and reporting about its operations as well as flaws in its internal risk control procedures. Price manipulation Boiler room: Selling/marketing penny stocks using improper conduct via creating a hpothetical market for thinly-traded products 27 Insider trading FT, 20 Nov 2011 The US Senate will this week consider a proposal to ban insider trading by lawmakers, spurred into action by increasing public outrage that representatives in Congress are allegedly benefiting from their positions and becoming richer at a time of national economic distress. (Financial Times Nov 20, 2011) Peter Schweizer (Stanford University s Hoover Institution), accuses Spencer Bachus, the Republican from Alabama who chairs the House financial services committee, of short selling stocks based on insider knowledge. Mr Schweizer found no less than 40 options trades by Mr Bachus between July and November of 2008, many of them within days of receiving closed door briefings from the likes of Henry Paulson, then Treasury secretary, and Ben Bernanke, chairman of the Federal Reserve. Read S.M. Bainbridge, 2011, «Insider trading inside the Beltway», Journal of Corporation Law. 28 Université Paris-Dauphine 14

15 Private information Information is costly Either because we pay for it Or because we produce it (mostly by analysis, comparison, verification and modeling of various public information) Private information is not in the possession of everyone Financial analysts are producers of private information 29 Financial analysts Financial analysts buy side (investment funds) Sell side (banks, brokers), published analyses are aggregated into consensus Collaborative sites ( ) Expert networks Independent analysts Brokerage commission with shared fees Recommendations, EPS revisions, consensus The added value of analysts? 30 Université Paris-Dauphine 15

16 Price Consensus ,5 6,0 5,5 5,0 4,5 4,0 EPS A consensus is computed by aggregating individual forecasts Most optimistic Consensus Avl Mai Price drop occurs prior to EPS revision Jun Jut Aût Sep Oct Nov Déc Jan Fév Mrs Avl Mai Jun Jut Aût Sep Oct Nov Déc Jan Fév Mrs 3,5 3,0 2,5 2,0 Most pessimistic Michelin's EPS forecast, Pric ,5 6,0 5,5 EP 1999 Consensus 50 5, Avl Mai Jun Jut Aût Sep Oct Nov Déc Jan Fév Mrs Avl Mai Price Price Jun Jut Aût 2001 Sep Oct Nov Déc Jan Fév Jut Aût Sep Oct Nov Déc Jan Fév Mrs Mrs 4,5 6,5 4,0 6,0 3,5 5,5 3,0 2,5 5,0 2,0 4,5 4,0 6,0 5,5 5,0 EPS EPS On a given date, forecasts are made for multiple fiscal years Michelin ,5 4,0 3,5 30 Jan Fév Mrs Avl Mai Jun Jut Aût Sep Oct Nov Déc Jan Fév Mrs , Université Paris-Dauphine 16

17 The added value of analysts Figure 20-5 Cumulative returns before and after analyst revisions, Ivkovic and Jegadeesh (2004). Upward (downward) revisions on the left (right) panel, based on 8,384 analysts working on 485 banks or brokers. The forecasts cover 3,184 to 6,182 companies per year. 33 The added value of analysts révision downward en baisse, revision du BPA en cours pas no revision de révision révision upward en revision hausse, du BPA en cours an 9 mois 6 mois 3 mois 1 mois 0 1 mois 3 mois 6 mois 9 mois 1 an avant révision après 1. EPS are updated after a price variation occurs 2. (very) small price variation after EPS is updated Grandin, Jacquillat (1991) 34 Université Paris-Dauphine 17

18 Event studies Objectives Quantify the direction and magnitude of price adjustments Quantify the speed of market reaction Approach Market reaction to new (unexpected) information Markets react to surprises. E.g. earnings announcement: the surprise is equal (?) to the difference between the announced earnings and the consensus of expected earnings The higher the number of observations, the better Allows to compute the distribution of abnormal returns : Hypothesis tests Excess or abnormal returns Modeling needed How to get rid of confounding events? (see below) Events are stacked relative to the announcement date (date 0) Implies that events must take place at different dates 35 "Abnormal" returns According to the CAPM The "normal" rate of return is E Ri, t = Rf + βi[ RM, t Rf And the "abnormal" return corresponds to the difference between the observed and the expected return, i.e. ε i, t = Ri, t Rf βi[ RM, t Rf "Market" model Normal return "abnormal" return "Mean" model Normal return "abnormal" return ] ( R i, t ) = RM t E, ( ) ] 36 Université Paris-Dauphine 18

19 Stacking Which event? Earnings announcement, director trades, IPO, public offering, share repurchases, change in rating, revision in analyst forecasts, etc. In all cases, Several events that impact different companies at different dates Stacking those events around the announcement date allows to build a portfolio of events and to eliminate the effects of confounding events If one is interested in the effect of hurricanes, catastrophic events (BP in Mexico), Lehman's bankruptcy, October 19, 1987 market crash, death of CEO, change in corporate laws As if only 1 event is available It is impossible to assess whether what is observed at that time is solely due to the event of interest 37 Speed of adjustment (of prices) to new information How to measure it? Event study Announcement by the US Federal Reserve CNBC morning call (analysts' views) Order imbalances (see chapter 3) Reversion to equilibrium after a shock The determinants of speed Frictions, delay and adjustment speed Behavioural issues? 38 Université Paris-Dauphine 19

20 How prices react to macro-economic news Ederington and Lee (1995) analyze the impact of 18 US Fed interest-rate announcements on fx futures trading prices over the period 1988 to The analysis is conducted through an event study that analyzes returns in 10 second intervals before and after the announcement. The authors classify the announcements as being positive/negative based on the direction of the first price change after the announcement. According to their results, most of the adjustment occurs during the first 40 seconds following the announcement and the total adjustment is completed in less than 2 minutes. Prices start to adjust 10 seconds after the announcement. The small magnitude of observed adjustments implies that potential gains (before costs!) are limited. 39 Price adjustment of futures contracts to macroeconomic announcements 0.08% 0.07% 0.06% 0.05% DEM EUR T-bond 0.04% 0.03% 0.02% 0.01% 0.00% -0.01% Ederington and Lee (1995) seconds 40 Université Paris-Dauphine 20

21 Adjustement to macro news Adjustment before announcement Suspicion of insider trading 18 Sep At 2 pm, the transactions skyrocket In NYSE, 400 million shares traded in 0.10 sec and 1 bn in 2 sec In Chicago (derivatives market), 5bn in 0.10 sec and 10 bn in 2 sec Transactions with high volumes satart simultaneously in Chicago and NY. Transmitting a signal from Washington to Chicago takes 7 miliseconds and shorter to NY Embargo principle (FOMC, Fed open market Committee) Journalists? Who has access to Press Room? What are the rules? What is the level of control? Vague regulation? Nanex : Can anyone start a news service and compete? How much are the news services paying? How much are they charging? Why was this not clearly disclosed? Why have a lock up room at all? 41 CNBC CNBC: «morning call» and«midday call» reports are among the most-watched TV shows in business. Analysts' viewpoints Event study over June 12 to October 27, stocks, 84 trading days Time 0 : time at which the name of the company is pronounced (or displayed) for the first time Speed: Adjustment takes less than 1 minute Asymmetric behavior (positive / negative report) Trading volumes Busse & Green, Université Paris-Dauphine 21

22 CNBC Stronger reaction to negative reports Morning neg Busse & Green, Adjustment after buy/sell imbalance Order imbalance (excess demand) is defined as the difference between buyerinitiated and seller-initiated trading volumes in dollars (OIB$). Sample: 20 largest NYSE companies in 1996 Chordia, Roll and Subrhamanyam (2001) R 50% of order imbalance on day t is explained by previous days' order imbalance (not displayed in the table) Returns computed as midpoint returns are explained both by return on the previous period (a1) and past (a3) and current (a2) order imbalances Beyond 15 minutes, past data fail to explain current returns M M t, t 1 = a1 Rt 1, t 2 + a2 OIB$ t, t 1 + a3 OIB$ t 1, t 2 a1 a a 2 3 From t to t-1 R² Coef. Test-T Coef. Test-T Coef. Test-T 5 minutes -0,066 (-7,40) 4,60 (20,1) 0,829 (5,45) 19,7% 10 minutes -0,095 (-15,8) 4,96 (11,6) 0,481 (4,44) 24,8% 15 minutes -0,083 (-9,89) 4,99 (11,2) 0,349 (3,35) 26,8% 30 minutes -0,028 (-2,27) 4,91 (10,7) 0,092 (0,97) 30,3% 1 hour -0,009 (-0,80) 4,95 (10,7) 0,007 (0,05) 32,8% 44 Université Paris-Dauphine 22

23 Delay, frictions and returns R = α + β R i, t i i M, t 4 4 ( n) ( n) δi RM, t n γ i Ri, t n ε i, t n= 1 n= R : return in excess of the risk-free rate Additional variables over CAPM measure market frictions and delay (MF) The above relationship is estimated annually in June using the most recent 52 1-week excess returns. Without frictions, coefficients on delay as well as contemporaneous correlations should be zero and the relationship reduces to the CAPM. The test is based on R 2 ratios of the two models (CAPM vs MF) D i 2 Ri, CAPM = 1 D=0, if additional variables have no 2 Ri, MF explanatory power 45 Significant relationship between delay and: trading activity number of analysts following the stock average number of daily trading prices available market value diversifiable risk book-to-market ratio Difference in highest vs lowest delay portfolios yields a 12% differential in returns Hou and Moskowitz (2005) Decile (based on D) D = 0,0008 0,0095 0,2474 0,3709 Average market value (million $) Book-to-market 0,6 1,0 3,3 9,0 Institutional investors (%) 51,6 42,6 9,2 6,3 Average trading activity (million $) 1 028,0 224,5 0,6 0,3 Diversifiable risk 9,5% 10,5% 15,5% 16,4% Number of analysts 22,2 13,5 1,4 1,4 Average price 129,1 43,1 6,8 4,7 Number of trading days 250,7 247,8 172,8 150,2 46 Université Paris-Dauphine 23

24 emini futures S&P500 futures, CME Indicators Michigan U. & Reuters : Index of Consumer Sentiment ISM (Arizona) subsidiary : Chicago PMI (purchasing managers index) Chronology (of a typical session) 9:30 Regular trading session opens for U.S. Stocks. 9:42 early release of Chicago PMI 9:45 release of Chicago PMI 9:53 early release of MCC (Michigan consumer confidence number) 9:55 release (Reuters) of MCC 10:00 scheduled economic news. Public release of MCC Reuters & early release 8 juillet 2013: Reuters abandons early-release 47 Market anomalies Definition Empirical evidence inconsistent with existing asset pricing theories Anomaly with respect to market efficiency are investors irrational? is "free lunch" possible? However, one should be cautious about potential profits without identifying the source of the anomaly? data mining is a real concern most of the time, anomalies persist after they have been discovered => hidden factor? 48 Université Paris-Dauphine 24

25 Market anomalies Anomaly Weekend effect January effect Friday the 13th effect Beginning of the month effect Holidays effect Halloween indicator Half-hour effect Religious holidays See Table 20-7 Evidence Calendar anomalies Monday returns are smaller Small caps exhibit higher returns in January In the US, returns on Fridays the 13th are lower than those on other Fridays Cumulated returns over the first half of a month account for more than 50% of the total return over the same month Returns are higher during the day that precedes a nonworking day Positive returns are mostly observed from November to May. One should sell stocks in May and buy-back them in November. A strong return over a particular 30-minute interval will repeat itself over the same interval on several trading days Specific return on days preceding a religious holiday (St Patrick, Rosh Hashanah, etc.). 49 Interpreting anomalies 1. They arise due to irrational behavior 2. They are simply statistical artefacts (data mining) Super-Bowl and Friday the 13th (to name but a few) 3. They prevailed but they have been arbitraged 4. They simply reflect a risk factor, Size effect, value effect 5. They are caused by market frictions 6. The anomaly does exist according to Fama-French (2008), but with no apparent justification? Excessive issuing price (IPOs), momentum effect, accrual effect 50 Université Paris-Dauphine 25

26 Size effect: the surviving "anomaly" Fact: returns are inversely related to size Small caps are more profitable. Holds especially on long horizon HOWEVER the difference in returns (small-caps - big-caps) is not constant over time and turns negative sometimes 51 The size effect on US markets Return 3% 2.5% 2% 1.5% 1% 0.5% 0 From 1973 to NYSE stocks 7659 Nasdaq stocks Small Size From Lamoureux and Sanger (1989) NYSE Nasdaq 52 Université Paris-Dauphine 26

27 Size premium since Difference in returns Small-Big (deciles) Reversed in 1928 and and and and and , The size effect is not an anomaly according to Fama-French (1992) 53 Interpreting the size effect Frictions, transaction costs, liquidity, information costs? All these variables are negatively correlated with size But how come these disadvantages have a timevarying effect? Reflection of a risk More profitable (on average) since riskier Less profitable sometimes which gives credit to the fact that some particular risk is at work small-caps are less robust in times of crisis (hence the Fama-French 3-factor model) 54 Université Paris-Dauphine 27