Portfolio Strategies Based on Analysts Consensus

Similar documents
Morningstar Investor Return

Market Timing with GEYR in Emerging Stock Market: The Evidence from Stock Exchange of Thailand

A Probabilistic Approach to Worst Case Scenarios

Strategic Decision Making in Portfolio Management with Goal Programming Model

Paul M. Sommers David U. Cha And Daniel P. Glatt. March 2010 MIDDLEBURY COLLEGE ECONOMICS DISCUSSION PAPER NO

Lifecycle Funds. T. Rowe Price Target Retirement Fund. Lifecycle Asset Allocation

Stock Return Expectations in the Credit Market

Idiosyncratic Volatility, Stock Returns and Economy Conditions: The Role of Idiosyncratic Volatility in the Australian Stock Market

The t-test. What We Will Cover in This Section. A Research Situation

Overreaction and Underreaction : - Evidence for the Portuguese Stock Market -

Capacity Utilization Metrics Revisited: Delay Weighting vs Demand Weighting. Mark Hansen Chieh-Yu Hsiao University of California, Berkeley 01/29/04

QUANTITATIVE FINANCE RESEARCH CENTRE. Optimal Time Series Momentum QUANTITATIVE FINANCE RESEARCH CENTRE QUANTITATIVE F INANCE RESEARCH CENTRE

Guidance Statement on Calculation Methodology

What the Puck? an exploration of Two-Dimensional collisions

What should investors know about the stability of momentum investing and its riskiness? The case of the Australian Security Exchange

Market timing and statistical arbitrage: Which market timing opportunities arise from equity price busts coinciding with recessions?

Local Does as Local Is: Information Content of the Geography of Individual Investors Common Stock Investments

AP Physics 1 Per. Unit 2 Homework. s av

Constructing Absolute Return Funds with ETFs: A Dynamic Risk-Budgeting Approach. July 2008

Homework 2. is unbiased if. Y is consistent if. c. in real life you typically get to sample many times.

Performance Attribution for Equity Portfolios

Economics 487. Homework #4 Solution Key Portfolio Calculations and the Markowitz Algorithm

Betting Against Beta

The Current Account as A Dynamic Portfolio Choice Problem

Bootstrapping Multilayer Neural Networks for Portfolio Construction

Using Rates of Change to Create a Graphical Model. LEARN ABOUT the Math. Create a speed versus time graph for Steve s walk to work.

296 Finance a úvěr-czech Journal of Economics and Finance, 64, 2014, no. 4

Sources of Over-Performance in Equity Markets: Mean Reversion, Common Trends and Herding

Rolling ADF Tests: Detecting Rational Bubbles in Greater China Stock Markets

Time-Variation in Diversification Benefits of Commodity, REITs, and TIPS 1

Evaluating the Performance of Forecasting Models for Portfolio Allocation Purposes with Generalized GRACH Method

Do Competitive Advantages Lead to Higher Future Rates of Return?

Measuring dynamics of risk and performance of sector indices on Zagreb Stock Exchange

ANALYSIS OF RELIABILITY, MAINTENANCE AND RISK BASED INSPECTION OF PRESSURE SAFETY VALVES

Smart Beta Multifactor Construction Methodology: Mixing versus Integrating

Momentum profits and time varying unsystematic risk

Centre for Investment Research Discussion Paper Series. Momentum Profits and Time-Varying Unsystematic Risk

A Liability Tracking Portfolio for Pension Fund Management

Interpreting Sinusoidal Functions

Monte Carlo simulation modelling of aircraft dispatch with known faults

Profitability of Momentum Strategies in Emerging Markets: Evidence from Nairobi Stock Exchange

KEY CONCEPTS AND PROCESS SKILLS. 1. An allele is one of the two or more forms of a gene present in a population. MATERIALS AND ADVANCE PREPARATION

Unsystematic Risk. Xiafei Li Cass Business School, City University. Joëlle Miffre Cass Business School, City University

Overview. Do white-tailed tailed and mule deer compete? Ecological Definitions (Birch 1957): Mule and white-tailed tailed deer potentially compete.

INSTRUCTIONS FOR USE. This file can only be used to produce a handout master:

Evaluating Portfolio Policies: A Duality Approach

Time & Distance SAKSHI If an object travels the same distance (D) with two different speeds S 1 taking different times t 1

MODEL SELECTION FOR VALUE-AT-RISK: UNIVARIATE AND MULTIVARIATE APPROACHES SANG JIN LEE

Proportional Reasoning

Asset Allocation with Higher Order Moments and Factor Models

An Alternative Mathematical Model for Oxygen Transfer Evaluation in Clean Water

Bill Turnblad, Community Development Director City of Stillwater Leif Garnass, PE, PTOE, Senior Associate Joe DeVore, Traffic Engineer

Machine Learning for Stock Selection

The Economic Costs of Vetoes: Evidence from NATO

CMA DiRECtions for ADMinistRAtion GRADE 6. California Modified Assessment. test Examiner and Proctor Responsibilities

Dynamics of market correlations: Taxonomy and portfolio analysis

Review of Economics & Finance Submitted on 27/03/2017 Article ID: Mackenzie D. Wood, and Jungho Baek

Can Optimized Portfolios Beat 1/N?

FIVE RISK FACTORS MODEL: PRICING SECTORAL PORTFOLIOS IN THE BRAZILIAN STOCK MARKET

Simulation based approach for measuring concentration risk

Portfolio Efficiency: Traditional Mean-Variance Analysis versus Linear Programming

Simulation Validation Methods

What is a Practical (ASTM C 618) SAI--Strength Activity Index for Fly Ashes that can be used to Proportion Concretes Containing Fly Ash?

A Study on the Powering Performance of Multi-Axes Propulsion Ships with Wing Pods

3. The amount to which $1,000 will grow in 5 years at a 6 percent annual interest rate compounded annually is

Examining the limitations for visual anglecar following models

Macro Sensitive Portfolio Strategies

As time goes by - Using time series based decision tree induction to analyze the behaviour of opponent players

Reliability Design Technology for Power Semiconductor Modules

Avoiding Component Failure in Industrial Refrigeration Systems

Revisiting the Growth of Hong Kong, Singapore, South Korea, and Taiwan, From the Perspective of a Neoclassical Model

Transit Priority Strategies for Multiple Routes Under Headway-Based Operations

A Stable Money Demand: Looking for the Right Monetary Aggregate

PRESSURE SENSOR TECHNICAL GUIDE INTRODUCTION FEATURES OF ELECTRIC PRESSURE SENSOR. Photoelectric. Sensor. Proximity Sensor. Inductive. Sensor.

DYNAMIC portfolio optimization is one of the important

Improving the Tournament Performance of ATP Players from the Perspective of Efficiency Enhancement

Testing Portfolio Efficiency with Non-Traded Assets: Taking into Account Labor Income, Housing and Liabilities

KINEMATICS IN ONE DIMENSION

AMURE PUBLICATIONS. Working Papers Series

WHO RIDE THE HIGH SPEED RAIL IN THE UNITED STATES THE ACELA EXPRESS CASE STUDY

2017 MCM/ICM Merging Area Designing Model for A Highway Toll Plaza Summary Sheet

The Measuring System for Estimation of Power of Wind Flow Generated by Train Movement and Its Experimental Testing

Valuing Volatility Spillovers

Urban public transport optimization by bus ways: a neural network-based methodology

SPECIAL WIRE ROPES The Value Line

2. JOMON WARE ROPE STYLES

The Effects of Systemic Risk on the Allocation between Value and Growth Portfolios

Keywords: overfishing, voluntary vessel buy back programs, backward bending supply curve, offshore fisheries in Taiwan

LSU RISK ASSESSMENT FORM Please read How to Complete a Risk Assessment before completion

Application of System Dynamics in Car-following Models

FHWA/IN/JTRP-2009/12. Panagiotis Ch. Anastasopoulos Fred L. Mannering John E. Haddock

TRACK PROCEDURES 2016 RACE DAY

Corresponding Author

WELCOME! PURPOSE OF WORKSHOP

BIOECONOMIC DYNAMIC MODELLING OF THE CHILEAN SOUTHERN DEMERSAL FISHERY

Slippery Slope? Assessing the Economic Impact of the 2002 Winter Olympic Games in Salt Lake City, Utah

Reexamining Sports-Sentiment Hypothesis: Microeconomic Evidences from Borsa Istanbul

Optimal Portfolio Strategy with Discounted Stochastic Cash Inflows

Towards a New Dynamic Measure of Competitive Balance: A Study Applied to Australia s Two Major Professional Football Leagues *

1. The value of the digit 4 in the number 42,780 is 10 times the value of the digit 4 in which number?

Transcription:

Porfolio Sraegies Based on Analyss Consensus Enrico Maria Cervellai Deparmen of Managemen Faculy of Economics Universiy of Bologna Piazza Scaravilli, 1 40126 Bologna Tel: +39 (0)51 2098087 Fax: +39 (0)51 237265 e-mail: cervellai@economia.unibo.i Firs draf: 1 March 2005 This draf: 1 Sepember 2005 Anonio Carlo Francesco Della Bina Deparmen of Managemen Faculy of Economics Universiy of Bologna Piazza Scaravilli, 1 40126 Bologna Tel: +39 (0)51 2098087 Fax: +39 (0)51 237265 e-mail: dellabina@economia.unibo.i, della@haas.berkeley.edu Pierpaolo Paioni e-mail: pierpaolopaioni@im.i Absrac Financial analyss research aciviy seems o be imporan for invesors in heir invesmen decisions. Undersanding if financial analyss repors can influence he marke and he degree of reliabiliy of heir forecass has been a heme lively debaed in he academic lieraure bu also in he press, mainly because of recen financial scandals. The main objecive of he paper is o calculae he invesmen value of financial analyss recommendaions on companies lised in he Ialian Sock Exchange and o verify he possibiliy of profiing from relying on he average consensus of recommendaions. We have enclosed in he analysis all he 16,634 repors issued beween he 1 s January 1999 and he 23 rd July 2004 and available on he websie of he Ialian Sock Exchange, consrucing a unique daabase for Ialy. Afer classifying companies by quarer, five porfolios are formed based on analyss average consensus o calculae he excess reurns of each porfolio in each quarer. Our resuls sugges ha analyss recommendaions have indeed invesmen value, even if invesors should carefully consider neural recommendaions ha can be considered as negaive ones. These resuls, furhermore, give some ineresing regulaory suggesions for a policy maker ha wans o ensure ransparency in he markes. Corresponding auhor. The auhors wan o hank Sandro Sandri, Marco Bigelli, Emanuele Bajo and Sefano Mengoli and he members of he Corporae Finance Group a he Deparmen of Managemen of he Universiy of Bologna for heir helpful commens and heir suppor. We are graeful o Robero Tasca for his suggesions and help in geing he daa on analyss recommendaions ha Emanuela Coni (R&D Office of Borsa Ialiana - Borsa Ialiana Group) provided us. We also hank Raghavendra Rau and Brad Barber for heir commens on a earlier draf of his paper, Sefano Bozzi, Umbero Cherubini, Sefano Fabrizio, Roland Gille and he paricipans o he Inernaional Finance Conference 2005 of he AFFI (French Finance Associaion) held in Paris, 27/28 June 2005, for furher suggesions on how o improve he paper. The usual disclaimer apply.

1. Inroducion Financial analyss research aciviy seems o be very imporan for invesors in deciding in which companies allocae heir wealh. This is mainly due by he fac ha gahering all he informaion necessary for invesmen decisions involves very high coss for a single unsophisicaed invesors. Undersanding if analyss repors can influence he marke and he degree of reliabiliy of heir forecass has been a heme lively debaed in he academic lieraure bu also in he press, mainly because of recen financial scandals. 1 The paper calculae he invesmen value of analyss recommendaions on companies lised in he Ialian Sock Exchange and verify he possibiliy of profiing from relying on he average consensus of recommendaions. We have enclosed in he analysis all he 16,634 repors issued beween he 1 s January 1999 and he 23 rd July 2004 and available on he websie of Borsa Ialiana. 2 Our daabase is unique, since i includes all he publicably available repors in he considered period. Following ar. 69 of he Consob 3 Regulaion on Issuers, in fac, all he repors issued by analyss mus be ransmied o he Consob and, simulaneously, o Borsa Ialiana ha publish hem. The archive of Borsa Ialiana can be accessed on a free basis and includes repors issued from January 1999. To verify he value of he recommendaions we have classified companies by quarer, based on he average consensus by analyss, and we have formed five porfolios based on his consensus. Furhermore, we have calculaed he excess reurns of each porfolio in each quarer. As far as we know, his is he firs paper ha proposes for Ialy such an analysis. The resuls seem o suppor he hypohesis of he invesmen value of a porfolio sraegy based on he average consensus of financial analyss. In he period considered, in fac, he porfolio ha includes he socks wih more favourable recommendaions records an average performance of 6.92% if calculaed wih he buy-and-hold reurns (BHR) mehodology and of 4.24% if we use he cumulaive abnormal reurn (CAR) mehodology, while he porfolio ha includes he socks wih less favourable recommendaions records a performance of 9.70% and 1 See, for example, he analysis of he Parmala case proposed by Ferrarini and Giudici (2005) and he implicaions in erms of reliabiliy of he informaion disseminaed by financial analyss. 2 Borsa Ialiana S.p.A. is he managing company of he Ialian Sock Exchange. 3 Consob (Commissione Nazionale per le Socieà e la Borsa) is he auhoriy responsible of supervising he Ialian Sock Exchange and he lised companies. 2

12.37% wih BHR and CAR respecively. The sraegy of an hypoheic invesor ha, following analyss recommendaions, buy he mos recommended socks and sell he less recommended ones, would yield abou he 16.6% (boh using he BHR and he CAR mehodologies). I is ineresing o noe he behavior of he porfolio ha only includes he sock ha receive neural recommendaions. Whereas, heoreically, his porfolio should record an abnormal reurn close o zero, empirically we find ha is performance is 2.27% wih BHR and 4.55% wih CAR. Invesors seem herefore o recognize he poenial conflic of ineress of financial analyss; in paricular when negaive recommendaions can damage he relaionships wih he covered company. The res of he paper is organized as follows: he second paragraph gives a survey of he lieraure, he hird presens he mehodology applied and he daabase used, he fourh commens he resuls obained and concludes. 2. Survey of he lieraure Several empirical sudies in he lieraure have focused on he predicing power of analyss, e.g. Diefenbach (1972), Bidwell (1977), Groh, Lewellen, Schlarbaum, Lease (1979), Copeland and Mayers (1982), Dimson and Marsh (1984). Womack (1996) analyzes a sample of 1,573 analyss recommendaion changes, issued beween 1989 and 1991, wih respec o 822 companies, lised in he US sock marke. 4 The analysis uses he informaion conained in he daabase of Firs Call Corporaion (now Thomson Financial), a company ha records in real ime all repors issued by analyss. The empirical evidence shows ha he socks subjec o a recommendaion change records an abnormal reurn significanly differen from zero: posiive (+ 2.4%) in case of upgrade, negaive (- 9.1%) in case of downgrade. 5 The asymmery beween he wo values can be explained wih he greaer frequency wih which analyss end o issue upgrades and wih he greaer cos of issuing a negaive repor. Several cases are known boh in he academic lieraure and in he financial press of analyss ha have been excluded from informaive meeing or ha have no received relevan informaion from he managemen of a company on which hey issued a negaive recommendaion. Thus, an analys face a rade-off beween he need of issuing repors ha are reliable, 4 Womack s work is subsequen o he sudy of Sickel (1995) ha is based on a sample of 17,000 changes of recommendaions issued by brokerage analyss beween 1988 and 1991. 5 The Cumulaive Abnormal Reurn (CAR) on a hree days window cenered on he even day and adjused for he size of he companies considered is 3% for buy recommendaions and - 4.7% for sell recommendaions. 3

o defend her repuaion, and he necessiy o mainain good relaionships wih he managemen of he covered companies. 6 The empirical resuls clearly show ha socks prices and volumes are influenced by recommendaion changes. The auhor highlighs ha analyss are paricularly good in sock picking bu also in marke iming, however hey mosly issue posiive repors and focus on companies wih higher capializaion. Barber, Lehavy, McNichols and Trueman (2001) assess he effecive profiabiliy of porfolio s sraegies based on he average consensus of analyss recommendaions. Whereas Womack s perspecive is analys-oriened and even-ime (e.g. o measure average price reacion o changes in analyss recommendaions), he perspecive of Barber e alii. is invesor-oriened and calendar-ime. In oher erms, while he firs sudy invesigaes he analyss and ime is measured wih he classical even sudy mehodology, he second one focuses on invesors and he analysis is performed in real calendar ime. This approach permis he auhors o direcly measure he abnormal gross reurns o a number of invesmen sraegies and o esimae porfolio urnover and he associaed ransacions coss incurred in implemening hem. The daa used in he paper come from he Zacks daabase for he period 1985 o 1996, which includes over 360,000 recommendaions from 269 brokerage houses and 4,340 analyss. For he sample period, Barber e alii. find ha buying he socks wih he mos favorable consensus recommendaions earns an annualized geomeric mean reurn of 18.8%, whereas buying hose wih he leas favorable consensus recommendaions earns only 5.78%. Afer conrolling for marke risk, size, book-o-marke, and price momenum effecs, a porfolio ha includes he mos highly recommended socks provides an average annual abnormal gross reurn of 4.13% while a porfolio of he leas favorably recommended ones yields an average annual abnormal gross reurn of 24.91%. Thus, purchasing he securiies in he op porfolio and selling shor hose in he lowes porfolio yields an average abnormal gross reurn of 75 basis poins per monh. In a subsequen research Barber e al. (2003) exend he sample period including 2000-2001, highligh ha he more highly recommended socks earned greaer markeadjused reurns during he 1996-1999 period han did he less highly favored socks. For he 2001-2000 period, he opposie is rue. The poor reurns of mos favored socks prevailed during mos monhs of 2000 and 2001 and characerized boh ech and nonech sock. The auhors found evidence consisen wih he possibiliy ha his reversal 6 See, he cases repored in Belcredi, Bozzi and Rigamoni (2003). 4

was a resul of analyss relucance o urn away from small-cap growh socks during his period, a ime when such socks significanly underperformed he marke. 7 The ecnique of consensus-based porfolios is also used by Boni and Womack (2003) wich examine he compeiion beween analyss. To add value o he recommendaions, analyss specialize in he sudy of few socks. The period considered is from 1996 o 2001. This work highlighs ha he reurns achievable hrough sraegies based on heir repors and on changes of recommendaions, record a Sharpe raio ha is five imes greaer han he one associaed wih a price momenum sraegy. In paricular, a sraegy consising in buying socks ha have been upgraded and selling socks ha have been downgraded is able o generae a monhly reurn of 1.4%, abou he 18% per year. Afer a monh from he change of recommendaion, he reurns from he socks recommended by analyss are posiive for 53 firms ou of 59. Analyss compeiion reduces he opporuniy o profi from changes of recommendaion: porfolios formed wih socks followed by a grea number of analyss generaes lower reurns. These resuls are also coheren wih he broad definiion of marke efficiency given by Grossman and Sigliz (1980), since posiive reurns are necessary o compensae for he coss needed o collec informaion. I seems, hus, ha analyss recommendaions have invesmen value o invesors. Using he heoreical framework proposed by Grossman and Sigliz, and wih regard o he Scandinavian counries, Von Nandeslsadh (2003) invesigaes he invesmen value in analys recommendaions. If he sock marke is efficien in he Grossman and Sigliz sense, hen invesors should no earn ne abnormal reurns by using analys recommendaions. In 1994-2001, he financial analys communiy s covered universe has ouperformed he corresponding marke index porfolio. The resuls show ha a sraegy based on he consensus has value for invesors and ha excluding from he analysis he recommendaions issued by invesmen banks he invesmen value grow even furher. Furhermore, he companies ha have received he greaes number of posiive recommendaions are generally characerized by high marke capializaion, inernaional coverage and marke-o-book raio as well as by a posiive rend of he prices in previous monhs. However aking ino accoun he ransacion coss arising from rading, he analysis does no find abnormal reurns ha are reliably differen from zero, unless we do no exclude from he sample he banks. 7 See also he recen research of Jegadeesh, Kim, Krische, Lee (2004). According o hese auhors framewok he level of he analyss consensus does no conain incremenal informaion when i is issued in correspondence wih oher predicive signals. I is he change in he analyss consensus, raher han he level, o be informaive. 5

3. The invesmen values of analyss recommendaions 3.1 Descripive analysis of he sample The daabase conains all he repors issued beween he 1 s January 1999 and he 23 rd July 2004 and available on he websie of Borsa Ialiana. However, we would like o highligh he fac ha a he end of July 2004, he archive on he websie conained abou 17,000 repors, 8 while he number of sudies received by Consob was abou 25,000. 9 There can be differen explanaion of his difference. The repors online, for insance, can be only a par of he repors available a Borsa Ialiana. 10 An alernaive explanaion is ha inermediaries send all he repors o Consob, bu only a par o Borsa Ialiana. Of course, his behavior would resul in conras wih he Consob Regulaion on Issuers. I would be desirable o solve his dilemma, and we believe ha Consob, as he auhoriy supervising he Ialian Sock Exchange should verify his anomaly and make available he resuls of his inquiry. 11 Saring from he whole sample, we have cleaned i eliminaing repors ha were no useful for our analysis, i.e. eliminaing double repors, non monographic, wihou any recommendaion or where i was ambiguous. The final sample includes by 12,791 repors issued by 68 financial inermediaries on 235 lised companies. In Appendix, we propose he main descripive saisics of he sample of repors wih recommendaion, ha consiues he saring poin of subsequen analysis. Comparing he number of repors received by each company wih is size, i is eviden ha analyss focus on bigger companies. From figure 1, i is clear ha he companies belonging o he firs quarile of capializaion received more han 57% of he repors, compared o a 7% of he las quarile. Table 1. Disribuion of repors per quarile of marke capializaion (size) Quarile of capializaion Average capializaion No. of repors Q1 9,776.4 6,716 Q2 836.15 2,804 Q3 187.56 1,353 Q4 46.69 821 8 Precisely, 16,634 repors. 9 In Consob, a he end of 2003 here were 21,032 repors, while a he end of 2004, 28,646. The arimeic average is 24,839, almos 25.000 repors herefore. Clearly, his is no he exac number of sudies received by Consob a he end of July 2004, since no necessarily he repors are issued uniformously during he year; however i can serve o compare wih he number of repors available in he Sock Exchange websie. 10 Emanuela Coni (R&D Office Borsa Ialiana - Borsa Ialiana Group), however, assured ha i seems ha only few sudies (abou 150) are available only in paper version. 11 To he bes of our knowledge here is no such clarificaion available. 6

We find suppor o he empirical evidence presened in he lieraure 12 ha financial analyss focus heir aenion on socks wih higher marke capializaion. A possible explanaion is ha analyss work more on big companies since hey are characerized by higher volumes of ransacions on which he financial inermediary for which hey work can earn higher rading and brokering commissions. Figure 1. Disribuion of repors per quarile of marke capializaion (size) Repors' number 7,000 6,000 5,000 4,000 3,000 2,000 1,000 0 6,716 2,804 1,353 821 Q1 Q2 Q3 Q4 Capializaion's quarile A simple analysis of he degree of correlaion beween he number of repors issued and he number of covered companies, highlighs ha a small number of inermediaries produce he greaes par of sudies, showing he an high degree of concenraion in he secor (see able 2). Table 2. Concenraion of marke shares Number of repors issued in he enire sample Firs inermediary 1,176 9.19% Firs wo inermediaries 2,332 18.23% Firs hree inermediaries 3,338 26.09% Firs four inermediaries 4,181 32.68% Firs five inermediaries 4,903 38.32% Firs en inermediaries 7,682 60.04% Remaining 58 inermediaries 5,109 39.96% Toal number of inermediaries 68 Toal number of sudies 12,791 100% 12 See Womack (1996) on he American marke. For Ialy, see Fabrizio (2000) and Cervellai, Della Bina (2004). 7

Furhermore, we highligh ha only few inermediaries cover mos of he companies, while he remaining prefer o focus jus on few lised companies. Comparing he number of repor issued by financial inermediaries wih he number of companies, i is clear ha he subjecs ha are more acive in issuing repors are also he ones ha cover he greaes number of companies (see able 3). This highligh he imporance of checking for poenial conflic of ineress of inermediaries ha have a relevan posiion in he research secor. Table 3. Number of repors issued vs number of covered companies, 1999-2004 No. of repors issued Degree of coverage (%) 50% ---40% 40% ---30% 30% ---20% 20% ---10% < 10% Toal Q1 7 5 2 0 0 14 Q2 0 1 6 6 0 13 Q3 0 0 0 9 5 14 Q4 0 0 0 0 13 13 Q5 0 0 0 0 14 14 Toal 7 6 8 15 32 68 Financial inermediaries are classified based on he number of monographic sudies issued and he number of covered companies. The ime period considered is beween Sepember 1999 and July 2004. The covered companies are hose wih a leas one recommendaion in each year. We consider 236 lised companies. The inermediaries are classified by row, based on he number of repors produced. The firs quinile (Q1) includes he 14 companies wih he higher number of repors issued in he period. The las quinile (Q5) includes insead he 14 companies wih he lower number of recommendaions issued in he enire period. The inermediaries are classified by colomn, based on he percenage class of covered companies. Percenages are expressed in erms of he oal number of covered companies in he enire period. 3.3 The recommendaions issued by financial analyss Analyss use a variey of sysems in heir recommendaions: five, six or hree poins scale, or even numeric sysems. For his reason i is difficul o compare raings issued by differen analyss. Since, furhermore, really few inermediaries repor he raing sysems hey use, i is necessary o pay aenion o compare recommendaions ha seem o be similar, bu ha are issued by analyss for differen financial inermediaries ha use differen raing sysem. In oher words, he same recommendaion could mean differen hings in differen raing sysems. To compare differen raing sysems i is 8

opporune o use an homogeneous scale. 13 We decided o use boh a hree-poins and a five-poins scale, o uniform our analysis o he prevailing inernaional lieraure in his field. The firs scale represens he simples ype of raing sysem since i divides he recommendaions in posiive, neural and negaive, using he raings buy, hold and sell. The second scale, he mos used in he lieraure and by analyss is insead a five-poins scale: buy, add, hold, reduce and sell. This raing sysem permis a wider range of raings adding a moderae posiive raing (add) and a moderae negaive judgmen (reduce). 14 Classifying he differen ypes of recommendaions wih respec o he chosen sysems, in ables 4 and 5 we presen he annual disribuion of recommendaions beween 1999 and 2004. Table 4. Annual disribuion of repors by ype of recommendaion (five poins scale) Year Buy Add Hold Reduce Sell Toal 1999 14 (0.11%) 28 (0.22%) 6 (0.05%) 3 (0.02%) 0 (0%) 51 (0.40%) 2000 514 (4.02%) 300 (2.35%) 337 (2.63%) 59 (0.46%) 49 (0.38%) 1,259 (9.84%) 2001 1,006 (7.86%) 547 (4.28%) 1,028 (8.04%) 182 (1.42%) 112 (0.88%) 2,875 (22.48%) 2002 966 (7.55%) 601 (4.70%) 1,011 (7.90%) 202 (1.58%) 74 (0.58%) 2,854 (22.31%) 2003 1,072 (8.38%) 1,034 (8.08%) 1,369 (10.70%) 270 (2.11%) 126 (0.99%) 3,871 (30.26%) 2004 584 (4.57%) 446 (3.49%) 698 (5.46%) 95 (0.74%) 58 (0.45%) 1,881 (14.71%) Toal 4,156 (32.49%) 2,956 (23.11%) 4,449 (34.78%) 811 (6.34%) 419 (3.28%) 12,791 (100%) Table 5 Annual disribuion of repors by ype of recommendaion (hree poins scale) Year Buy Hold Sell Toal 1999 42 6 3 51 (82.35%) (11.76%) (5.88%) (0.40%) 2000 814 337 108 1,259 (64.65%) (26.77%) (8.58%) (9.84%) 2001 1,553 1,028 294 2,875 (54.02%) (35.76%) (10.23%) (22.48%) 2002 1,567 1,011 276 2,854 (54.91%) (35.42%) (9.67%) (22.31%) 2003 2,106 1,369 396 3,871 (54.40%) (35.37%) (10.23%) (30.26%) 2004 1,030 698 153 1,881 (54.76%) (37.11%) (8.13%) (14.71%) Toal 7,112 4,449 1,230 12,791 (55.60%) (34.78%) (9.62%) (100%) 13 In Ialy, Belcredi, Bozzi and Rigamoni (2003) use a eigh-poins scale, while Fabrizio (2001) a four-poins scale. 14 Someimes, he erms add and reduce, are used as synonimous of ouperform and underperform. See Cervellai, Della Bina and Giulianelli (2005). 9

From he above figures i is possible o noe how he reporing aciviy of analyss is increased in he las years, 15 and he percenage of posiive recommendaions is always greaer han he percenage of negaive ones. Considering able 5, in fac, i is eviden ha, in he whole period considered, 7,112 repors (55.6% of he oal) repor a posiive recommendaion, while only 1,230 (9.62% of he oal) a negaive one. This evidence is well known and debaed in he lieraure. Usually researchers have advanced wo main explanaions: analyss excessive opimism or conflic of ineress. The firs hypohesis refers o he fac ha analyss seem o be oo opimisic on he perspecives of he socks hey follow. The second one argues ha analyss prefer no issuing any repor insead of issuing a negaive one. Classifying he recommendaions based on he curren and previous raing i is possible o form a marix of he changes of recommendaions, highlighing he frequency of upgrades and downgrades. As i is possible o see from ables 6 and 7, he greaes par of he repors does no conain changes of recommendaion: in he five-poins scale he unchanged repors are he 84.06%, while in he hree-poins scale are he 86.94%. Table 6 Summary able of changes of recommendaions (five- poins scale) Changes of recommendaion Number of repors Percenage Unchanged 9,253 84.06% Upgrade 851 7.73% Downgrade 904 8.21% Toal 11,008 100% Table 7. Summary able of changes of recommendaion (hree-poins scale) Changes of recommendaion Number of repors Percenage Unchanged 9,570 86.94% Upgrade 687 6.24% Downgrade 751 6.82% Toal 11,008 100% No every repor conains he previous raing, since some of hem are iniiaion of coverage and ohers are preceded by repors in which here is no recommendaion. Considering his fac, he sample size reduces o 11,008 repors for which we also have he previous raing. Table 8 presens he selecion crieria of he repors wih curren and previous raing, ha consiues he basis for consruc he marices of changes of recommendaions. 15 In 2004 he number decreases, bu his is due o he fac ha we have repors jus unil he 23 rd July. 10

Table 8. Summary able of repors wih curren and previous raings Toal number of monographic sudies 16,634 100% Sudies ha are double, wihou raing, wih ambiguous raing (3,843) (23.10%) Toal number of monographic sudies wih raing 12,791 76.90% Sudies wihou previous raing (1,235) (7.42%) Toal number of monographic sudies wih previous raing 11,556 69.47% Sudies wihou curren raing 16 (548) (3.29%) Toal number of monographic sudies wih previous and curren raing ha form he sample of observaions 11,008 66.18% Observing he marices of he changes of recomendaions, i is possibile o noe ha he greaes par of unchanged posiions is referred o, in a five-poins sysem (able 9), o buy (28.98%), add (18.85%) and hold (29.51%) recommendaions; and in a hreepoins sysem (able 10) o buy (50.50%) and hold (29.51%) recommendaions. The percenage of upgrades, furhermore, is less, even if slighly, o ha of downgrades. Considering he five-poins scale, he upgrades are he 7.73%, while he downgrades are he 8.21%. Wih reference o he hree-poins scale, he upgrades are only he 6.24%, while he downgrades are he 6.82%. Table 9. Marix of changes of recommendaion (five poins scale) Curr en Rain g Buy Add Hold Reduce Sell Toal Previous Raing Buy Add Hold Reduce Sell Toal 3,190 152 198 14 5 3,559 (28.98%) (1.38%) (1.80%) (0.13%) (0.05%) (32.33%) 142 2,075 280 47 1 2,545 (1.29%) (18.85%) (2.54%) (0.43%) (0.01%) (23.12%) 239 274 3,248 111 31 3,903 (2.17%) (2.49%) (29.51%) (1.01%) (0.28%) (35.46%) 20 30 130 507 12 699 (0.18%) (0.27%) (1.18%) (4.61%) (0.11%) (6.35%) 5 2 51 11 233 302 (0.05%) (0.02%) (0.46%) (0.10%) (2.12%) (2.74%) 3,596 2,533 3,907 690 282 11,008 (32.67%) (23.01%) (35.49%) (6.27%) (2.56%) (100%) 16 In his caegory we consider monographic sudies for which i does exis a previous raing, bu ha have no been included in he marices of changes of recommendaion since, for example, he valuaion of he sock has been emporarily suspended or since he analys ha iniially covered he sock is changed. 11

Table 10. Marix of changes of recommendaion (hree poins scale); percenages in parenheses Previous Raing Cur ren Rai ng Buy Hold Sell Toal Buy Hold Sell Toal 5,559 478 67 6,104 (50.50%) (4.34%) (0.61%) (55.45%) 513 3,248 142 3,903 (4.66%) (29.51%) (1.29%) (35.46%) 57 181 763 1,001 (0.52%) (1.64%) (6.93%) (9.09%) 6,129 3,907 972 11,008 (55.68%) (35.49%) (8.83%) (100%) This resul seems in conras wih he previous sudies in he lieraure, bu i is possibile o explain i considering he fac ha he greaes par of he period considered refers o bear markes. I is, furhermore, coheren wih he hypohesis of overopimism of he analyss. 17 3.4 Mehodology In his paragraph we describe he mehodology used o deermine he value of an invesmen sraegy based on he average consensus of analyss recommendaions. As a firs sep, we have calculaed, for each period and company, he average consensus. As ime period of reference we have chosen he quarer. The reason is wofold: on a pracical ground, o have enough recommendaions in each porfolio in every period, we could no use a monhly basis as i has been used in oher sudies in he lieraure; 18 from a heoreical poin of view we argue ha he quarer consiues for many porfolio managers he righ period for performance evaluaion and porfolio rebalancing, more ofen if he invesmen is managed hrough banks or muual funds. To deermine he average consensus on a company, in a given quarer, i has been necessary o aribue a numeric value o each raing. The scale ha we have used is he following: Buy = 1; Add = 2; Hold = 3; Reduce = 4; Sell = 5. The average consensus per quarer for a company is calculaed as he sum of all he raings issued by analyss on ha company in he quarer, and diving by he number of repors in he same period. Formally: A i, = n 1 i, n i, j = 1 A i, j, 17 See Cervellai, Della Bina, Giulianelli (2005). 18 See Barber e al. (2001). 12

where: A, is he average consensus on company i in quarer ; i A i j,, is he individual raings conained in each of he n i, repors issued in he quarer on he considered sock; n i, is he number of repors issued on sock i in quarer. The average consensus hus calculaed, however, does no allow o have an idea of he degree of agreemen or disagreemen among analyss ha have issued raings on he considered company. We have, herefore, decided o inroduce a simple measure of dispersion of he recommendaions around he average consensus. As a measure of dispersion we have used he sandard deviaion: D i, = ni, ( A j = 1 i, j, n i, A i, ) 2 where: D i, dispersion level, for quarer, around he average consensus; A, average i consensus on company i in quarer ; A i j,, is he individual raings conained in each of he n i, repors issued in he quarer on he considered sock; n i, is he number of repors issued on sock i in quarer. Once classified he companies following he average consensus in each quarer, i is possible o form porfolios based on his consensus. Five porfolios have been formed, for each quarer: porfolio 1: companies wih he highes raings, i.e. hose wih average consensus in beween 1 and 1.5 porfolio 2: companies wih posiive raings, i.e. hose wih average consensus in beween 1,5 and 2,5 porfolio 3: companies wih an inermediae consensus, i.e. hose wih consensus beween 2,5 and 3,5 porfolio 4: companies wih a sligh negaive consensus, i.e. hose wih consensus beween 3,5 and 4,5 porfolio 5: companies wih a very negaive consensus, i.e. hose wih consensus beween 4,5 and 5 To evaluae he performances for every quarer of hese porfolios we have used wo disinc mehodologies: CAR (Cumulaive Abnormal Reurn) and BHR (Buy and Hold Reurn). CAR mehodology consiss in summing he excess reurns recorded in he considered period. More formally: 19 ( ) CAR i, s = T ( Ri, E Ri, ) = 1 where: CAR i,s is he cumulae abnormal reurn of company i in quarer s ; R i, is he reurn of company i in day ; E(R i, ) is he expeced reurn of company i in day 19 See Barber, Lyon (1997) and Lyon, Barber,Tsai (1999). 13

. The difference R i, - E(R i, ) represens, herefore, he abnormal reurn of company i in day. Once obained he CAR for every company, we have compuer CARs for he porfolios as an average of he CARs of he companies in each porfolio for every quarer. 20 More formally, he porfolio CAR in each quarer is given by: CAR p, s n CAR i, s i= = 1 n where: CAR p,s is he abnormal reurn of porfolio p in quarer s ; n is he number of socks forming he porfolio p. A limiaion of he CAR mehodology, however, is ha i assumes ha one should periodically adjus he porfolio o equally disribue he wealh invesed in he porfolio among differen socks. Using he BHR mehodology, he reurn in each quarer of a sock is expressed as: BHR T T ( 1+ Ri, ) ( + E ( Ri, ) ) i, s = 1 = 1 = 1 where: BHR i,s is he excess reurn of sock i in quarer s ; R i, is he reurn of sock i in day ; E(R i, ) is he expeced reurn of sock i in day. The porfolio BHR is jus he average of single socks BHRs: BHR p, s n i = = 1 BHR n i, s where: BHR p,s is he excess reurn of porfolio p in quarer s ; n is he number of socks in porfolio p. Following Barber and Lyon (1997) we consider as an esimae of he expeced reurn E(R i, ) he reurn of a marke index R m,. 21 Boh mehodologies can herefore be re-expressed wih he following formulas: CAR i, s = T ( Ri, Rm, ) = 1 BHR T T ( 1+ Ri, ) ( + Rm, ) i, s = 1 = 1 = 1 20 The underlying assumpion here is ha he oal amoun invesed in each porfolio is equally divided among all he socks. 21 The marke index used here is he Mibel (Milano Indice Borsa Telemaica), a global index represening he general rend of he socks lised in he Ialian Sock Exchange. 14

To calculae daily reurns for individual socks we have decided, following he main conribuion in he lieraure, o use differen mehods for CAR and BHR. For CAR we have used a coninuously compounded reurn 22, whereas for BHR a discree compounded reurn. The wo mehodologies can be expressed as follows: P R = ln P 1 P P R = P 1 1 where: P sock price in day ; P -1 sock price in day -1. To es he null hypohesis ha he reurns calculaed wih BHR or CAR are equal o zero for he sub-sample of n companies forming he porfolio, we use he following parameric ess: 23 CAR = CAR i ( σ ( CAR ) n ) i BHR = BHR i ( σ ( BHR ) n ) i where: CAR i and BHR i are he porfolio average excess reurns; σ(car i ) and σ(bhr i ) are he sandard deviaions of abnormal reurns for he n socks forming porfolio i. 3.5 The invesmen value of analyss recommendaions In his paragraph we examine he invesmen value of a sraegy based on he average consensus of analyss recommendaions. For each porfolio and every quarer, we have deermined he average consensus of raings issued on each sock from analyss ha have ousanding recommendaions on ha sock in he considered quarer.we have also calculaed excess reurns, adjused by he marke reurns, using CAR and BHR mehodologies. If analyss recommendaions have value, ordering he porfolios from 1 o 5, wih 1 represening he porfolio conaining he bes raings and 5 he porfolio conaining he wors ones, we would expec o observe he following effecs: porfolio 1 should have he mos posiive adjused excess reurn; porfolio 2 should have a posiive excess reurn, bu lower han porfolio 1; porfolio 3 should have adjused excess reurns close o zero; 22 Using he coninuously compounded reurn one assumes ha P = P -1e R, where R is he rae of reurn during he period ( 1, ). 23 See Barber and Lyon (1997). 15

porfolio 4 should have a negaive adjused excess reurn; porfolio 5 should have he mos negaive adjused excess reurn. Table 11 shows he oal reurn in each quarer for every porfolio. The reurns are all saisically significan and reflec he expecaions, confirming our hypohesis. The Average Dispersion, repored in second column of able 11, measures he degree of agreemen beween analyss wihin each class of raing. By consrucion, i represens he sandard deviaion, adding addiional informaion wih respec o he mere average consensus in he class. I would be possible o have he same average consensus, bu a differen dispersion and, herefore, a raher differen degree of agreemen. Table 11. Summary resuls for every porfolio in quarer Porfolio Average Toal number of [Class] Dispersion repors BHR() -Sa CAR() -Sa Porfolio 1 [1 -- 1.5] 0.21 1,942 6.92% 4.5033*** 4.24% 5.3385*** Porfolio 2 [1.5 -- 2.5] 0.70 6,898 2.01% 3.6426*** 0.55% 1.0205 Porfolio 3 [2.5 -- 3.5] 0.39 3,366-2.27% 3.1332*** -4.55% 6.1156*** Porfolio 4 [3.5 -- 4.5] 0.52 531-5.29% 2.3631** -8.56% 3.9064*** Porfolio 5 [4.5 -- 5] 0.00 54-9.70% 3.2324*** -12.37% 3.5922*** Saisical significance : * = 10%, ** = 5%, *** = 1% Wih reference o able 11, ake for example porfolio 1, ha presens a low average dispersion, equal o 0.21. However, i should be considered ha he range of he class is only 0.5, from 1 o 1.5, herefore he incidence of he average dispersion is 42%. Porfolio 2 is he one in which single raings are more dispersed, 70%, followed by porfolios 3 and 4 wih, respecively, 52% and 39%. Porfolio 5 has no dispersion. Table 11 conains oher ineresing resuls. Considering he number of repors in each porfolio, i is eviden ha he porfolios ha have he greaer number of repors are hose conaining non-negaive raings. This resul can be addressed using differen explanaion. The firs one suppors he hypohesis of an opimisic bias of analyss ha end o view he socks ha hey follow oo favorably. 24 The second hypohesis claims ha analyss are relucan o issue negaive raings, o avoid problems wih he managemen of he covered companies. A hird explanaion can be ha analyss simply follow, on average, socks 24 This explanaion is proposed by he behavioral approach o finance, ha relae psicology and finance. 16

wih beer performances. Figure 2 clearly show ha he adjused reurns of he five porfolios are in line, boh considering he BHR and he CAR mehodology, wih he level of average consensus of analyss recommendaions. This seems o confirm he invesmen value of a sraegy based on analyss consensus. Figure 2. Toal average reurn compued for every porfolio in quarer BHR vs CAR 10.00% 5.00% Reurn 0.00% -5.00% -10.00% BHR() CAR() -15.00% 1 2 3 4 5 Porfolio Porfolio 1 has recorded an average reurn of 6.92%, wih reference o all he quarers considered, using BHR and 4.24% wih CAR. Porfolio 5, insead, had a performance of 9.7% wih BHR and 12.37% wih CAR. Adoping a porfolio sraegy based on he consensus of financial analyss, i.e. buying he socks wih he more favorable recommendaions and selling he leas recommended ones, an invesor could gain an abnormal reur of abou 16.6%, boh wih BHR and CAR, as highlighed in able 12 ha conains he differences beween quarerly average reurns. I seems, herefore, ha analyss recommendaions have real invesmen value for invesors. However, he presen analysis does no ake ino accoun ransacion coss. I is necessary o ake ino accoun he commissions, as well as he bid-ask spread and he oher coss relaed o ransacions o calculae he ne reurn for invesors. The paper by Barber e al (2001) shows, in fac, ha aking ino accoun hese coss, he abnormal reurns recorded following analyss recommendaions end o disappear. Von Nandeslsadh (2003), wih reference o Scandinavian markes, finds no abnormal 17

reurns, once ha i akes ino accoun ransacion coss. I is ineresing, however, o highligh ha invesors can obain posiive abnormal reurns if hey follow only he recommendaions of analyss ha do no work for a bank. The resuls for porfolios 2 and 4 seem o be in line wih expecaions as well. The former has recorded an average reurn of 2.01% wih BHR and 0.55% wih CAR, 25 while he laer has realized 5.29% wih BHR and 8.56% wih CAR. The resuls for Porfolio 3 are insead somehow surprising, or a leas of difficul inerpreaion. This porfolio should heoreically have an excess reurn close o zero, while for our sample i recorded a 2.27% wih BHR and a 4.55% wih CAR. A possible explanaion of hese negaive reurns can be advanced referring o he incenives ha analyss have o issue a neural raing, insead of a negaive one. Several sudies in he lieraure, bu also aricles in he financial press, have shown ha analyss can face several problems afer issuing a negaive recommendaion. There have been cases in which analyss have been excluded from meeings wih he managers of a company or from receiving relevan informaion afer having issued a negaive recommendaion. Analyss herefore face a rade-off beween issuing correc raings, o build heir own repuaion, and mainain good relaionships wih he managemen of he companies hey follow o have access o he necessary informaion hey need for heir research aciviy. I seems ha his rade-off pushes analys o be upward biased, i.e. he endency o issue neural raings while insead hey should issue negaive ones, or even no o issue negaive repors a all. 26 In able 13 we provide he differences beween annual reurns on he five porfolios, for BHRs and CARs, referred o he whole period considered. The differences in erms of annual reurns among porfolios are of relevance and saisically significan, in paricular he difference beween exremes (porfolio 1 and 5) is large (16.61% wih BHR and 16.60% wih CAR) and significan a he 1% confidence level. I is worh o noice ha only he differences beween porfolios 3 and 4 for BHR and beween porfolios 4 and 5 for CAR are no saisically significan, probably due o he fac, already discussed in he paper and in lieraure, ha hold and reduce raings can be considered as negaive recommendaions, no significanly differen from sell raings. 25 This value (0.55%), however, i is no saisically significan. 26 The bias induced by omiing o issue negaive repors is well illusraed by Fabrizio (2001). 18

Tab. 12. Differences beween average quarerly reurns for each porfolio (-Sa in brackes) Pare A. Comparison beween he five porfolios using average BHR Porfolio 1 [1 -- 1.5] Porfolio 2 [1.5 -- 2.5] Porfolio 3 [2.5 -- 3.5] Porfolio 4 [3.5 -- 4.5] Porfolio 5 [4.5 -- 5] Porfolio 1 [1 -- 1.5] Porfolio 2 [1.5 -- 2.5] Porfolio 3 [2.5 -- 3.5] Porfolio 4 [3.5 -- 4.5] Porfolio 5 [4.5 -- 5] 0.00% - - - - 4.91% (3.0069)*** 9.19% (5.4101)*** 12.21% (4.4967)*** 16.61% (4.9297)*** 0.00% - - - 4.28% (4.6990)*** 7.30% (3.16630)*** 11.71% (3.8380)*** 0.00% - - 3.02% (-1.2817) 7.42% (2.4052)** 0.00% - 4.41% (-1.1777) 0.00% Pare B. Comparison beween he five porfolios using average CAR Porfolio 1 [1 -- 1.5] Porfolio 2 [1.5 -- 2.5] Porfolio 3 [2.5 -- 3.5] Porfolio 4 [3.5 -- 4.5] Porfolio 5 [4.5 -- 5] Porfolio 1 [1 -- 1.5] Porfolio 2 [1.5 -- 2.5] Porfolio 3 [2.5 -- 3.5] Porfolio 4 [3.5 -- 4.5] Porfolio 5 [4.5 -- 5] 0.00% - - - - 3.68% (3.8273)*** 8.78% (8.0774)*** 12.80% (5.4904)*** 16.60% (4.6991)*** 0.00% - - - 5.10% (5.5398)*** 9.12% (4.0372)*** 12.92% (3.7074)*** 0.00% - - 4.01% (1.7338)* 7.82% (2.2201)** 0.00% - 3.81% (0.9331) 0.00% Saisical significance : * = 10%, ** = 5%, *** = 1% To es he hypohesis ha neural raings can be associaed wih negaive judgmens by he analyss, we furhermore divide he socks in wo porfolios, respecively formed by companies wih non-negaive consensus (1 -- 3), including he hold raings as nonnegaive, and he ones wih negaive average consensus (3 -- 5). Observing able 13, i is possible o noe ha he firs porfolio, wih an average dispersion of he raings equal o 0.45, has recorded a posiive average excess reurn of 2.10% wih BHR and 0.13% wih CAR, 27 while he second porfolio has realized, wih an average dispersion of raings of 0.57, a negaive excess reurn (-4.46% wih BHR and 7.73% wih CAR). 27 This value (CAR= 0.13%), however, is no saisically significan. 19

Table 13. Toal average reurn per quarer calculaed dividing among sock wih non negaive recommendaions (1 -- 3) and wih negaive ones ( 3 -- 5 ) Class of Average Toal number Recommendaions Dispersion of repors BHR() -Sa CAR() -Sa Non negaive [1 -- 3] 0,45 11,463 2.10% 3.8964*** 0.13% 0.31586 Negaive [3 -- 5] 0,57 1,328-4.46% 3.0099*** -7.73% 5.2196*** Saisical significance: * = 10%, ** = 5%, *** = 1% Figure 3. Toal average reurn per quarer calculaed dividing among sock wih non negaive recommendaions (1 -- 3) and wih negaive ones ( 3 -- 5 ) BHR() BHR vs CAR CAR() 4.00% 2.00% R e u r n 0.00% -2.00% -4.00% -6.00% -8.00% Non Negaive Negaive Porfolio Noe: he raing hold is included in he porfolio of Non Negaive recommendaions (1 -- 3) Coherengly wih he hypohesis ha neural recommendaions should be considered as negaive raings, in able 14 and figure 4 we presen an alernaive classificaion. The firs porfolio now includes only sric posiive raings ([1 -- 3]), excluding neural recommendaions, while he second porfolio includes non posiive raings ([3 -- 5]), including his ime he hold raing. Table 14. Toal average reurn per quarer calculaed dividing among socks wih posiive recommendaions (1 -- 3) and wih non posiive ones (3 -- 5) Class of Average Toal number Recommendaions Dispersion of repors BHR() -Sa CAR() -Sa Posiive [1 -- 3] 0,54 10,593 3.37% 5.3497*** 1.42% 3.28006*** Non Posiive [3 -- 5] 0,29 2,198-3.48% 4.3206*** -6.00% 7.39231*** Saisical significance : * = 10%, ** = 5%, *** = 1% 20

Figure 4. Toal average reurn per quarer calculaed dividing among sock wih posiive recommendaions (1 -- 3) and wih non posiive (3 -- 5). BHR vs CAR BHR() CAR() 4.00% 2.00% R e u r n 0.00% -2.00% -4.00% -6.00% -8.00% Posiive Porfolio Non posiive Noe: he raing hold is included in he porfolio of Non posiive recommendaions (3 -- 5) In his alernaive definiion, he firs porfolio records an adjused reurn of 3.37% wih BHR and 1.42% wih CAR, wih an average dispersion of raings of 0.54; whereas he second porfolio realizes a performance of -3.48% wih BHR and -6% wih CAR, wih an average dispersion of 0.29. We should highligh ha he posiive reurns associaed wih he Posiive raings porfolio are higher han before. A he same ime he reurns of he Non Posiive porfolio are beer han before, since now we have added he neural recommendaions o his second porfolio and eliminaed from he firs one. Firs of all, we should noe ha he number of repors in he second porfolio increases, wih benefis in erms of robusness of he resuls. In fac, whereas in he firs classificaion (non negaive vs negaive) he CAR was no saisically significan, in his alernaive definiion, no only is significanly differen from zero, bu i is also higher in magniude. Furhermore, while he average dispersion in he posiive porfolio almos remains he same, he one associaed wih he second porfolio dramaically decreases, 28 suggesing an higher degree of agreemen beween non posiive raings, once neural and negaive recommendaions are pooled ogeher. 28 I slighly increases in absolue erms for he posiive porfolio, bu since he range of raings narrows, in relaive erms i decreases. For he non posiive porfolio he decrease is even bigger if we consider he wider range of raings ha are now included in he non posiive porfolio. 21

Conclusions The paper examines he possibiliy of profiing from an invesmen sraegy based on he average consensus of analyss recommendaions. If on one hand individual and insiuional invesors can be willing o bear he cos for analyss repors, on he oher hand marke efficiency ells us ha hose repors should have no value. Therefore, i remains o be verified if analyss recommendaions have or no invesmen value. We have hen creaed a daabase including he recommendaions issued by analyss in monographic sudies issued beween he 1 s January 1999 and he 23 rd July 2004 and publicly available on he websie of he Ialian Sock Exchange. Firs of all, we have performed a descripive analysis of he sample, highlighing some ineresing feaures of he reporing aciviy in he Ialian sock marke. From a comparison beween he number of repors received from each company and is size, we have shown ha analyss prefer o issue repors on bigger companies. An explanaion of his phenomenon is ha since bigger companies are characerized by an higher number of ransacions, hey could allow for some economic benefis deriving from he commissions on rading and brokering aciviy. Few inermediaries produce he majoriy of repors and he more acive in issuing sudies are also he ones covering he majoriy of firms. This evidence highlighs he imporance of conrolling for poenial conflic of ineress of inermediaries ha have a relevan posiion in he marke of repors. Considering he disribuion of recommendaions issued by analyss, we have also shown ha he percenage of posiive raings is always greaer ha he fracion of negaive ones. This evidence can be explained in wo alernaive ways: analyss can show excessive opimism in heir reporing aciviy, or hey can jus omi o issue a negaive repor o avoid problems wih he managemen of he companies, ha is he main source of he informaion hey use. Apar from his preliminary and descripive analysis of he sample, o verify if analyss repors have any invesmen value, we have formed five porfolios, dividing socks on he base of he average consensus for each quarer of he sample. We used he CAR and BHR mehodologies o calculae average abnormal reurns of he five porfolios for each quarer and for he period as a whole. Comparing excess reurns of each porfolio in he enire period of ime ha we have considered wih he level of 22

average consensus of analyss raing, we found resuls in line wih our inuiion. Porfolio formed by very or moderaely posiive raings record a posiive excess reurn, while porfolios wih very or moderaely negaive raings have shown negaive excess reurns. The porfolio conaining neural raings gives insead ambiguous resuls. From a heoreical poin of view i should record excess reurns close o zero. The resuls, insead, show negaive excessive reurns boh wih he CAR and BHR mehodologies. An explanaion, well-acceped in lieraure, is ha neural raings can be considered as negaive ones, since in general analyss end o issue very few reduce or sell raings. Afer having performed he proper ess for saisically significance, we find ha analyss recommendaions have indeed invesmen value if we consider an horizon ha is a leas annual, or ha ake ino consideraion he whole sample. The resuls shown hus far have no considered ransacion coss. We should include hese coss in he analysis o see if analyss recommendaions really convey invesmen value or if hey, even if posiive, would no be sufficien o cover hose coss. Fuure research will have o consider his aspec. More generally, however, we can conclude ha seems ha invesors can rely on analyss average consensus, wih a cauion, o consider very carefully neural recommendaions ha, as shown in he lieraure, can be considered as negaive ones. The reporing aciviy seems herefore o significanly influence he invesmen decisions of invesors, and under his ligh i can be seen he increasing amoun of regulaion of he Ialian and European legislaors. The main objecive of hese regulaions on reporing aciviy is in fac o favor he diffusion of ransparen and imely relevan and price sensiive informaion o help invesors in heir decisions. In his regard, we argue ha legislaors should impose more precise crieria on he more delicae aspec conained in he recommendaions, i.e. he neural raing. If one objecive of regulaion is o enhance ransparency and disclosure, hen i is necessary ha invesors really undersand he meaning of every recommendaion. On his respec, i seems appropriae o quoe Irving Fisher: Risk is inversely proporional o knowledge. 23

References Barber B., Lehavy R., McNichols M., and Trueman B. (2001) Can invesors profi from he prophes? Securiy analys recommendaions and sock reurns, The Journal of Finance, Vol.56, pp. 531-563. Barber B., Lehavy R., McNichols M., and Trueman B. (2003) Reassessing he reurns o analyss' sock recommendaions, Financial Analyss Journal,Vol. 59, 2, pp. 88-96. Barber B., Lyon J.D (1997) Deecing long run abnormal sock reurns: he empirical power and specificaion of es saisic, Journal of Financial Economics, Vol. 43, pp. 341-372. Belcredi M., Bozzi S., Rigamoni S. (2003) The Impac of Research Repors on Sock Prices in Ialy, working paper, Universià Caolica del S.Cuore, Universià Luiss G.Carli. Bidwell C. (1977) How good is brokerage research?, Journal of Porfolio Managemen, 3, pp. 26 31. Boni L., Womack K. (2003) Analyss, Indusries, and Price Momenum, working paper, Universiy of New Mexico, Darmouh College. Cervellai E.M, Della Bina A. (2004) Analisi finanziari: conflii d'ineresse o eccessivo oimismo. Evidenza empirica dal mercao ialiano delle IPO, Banca Impresa e Socieà, n.2, pp. 367-399. Copeland T., Mayers D. (1982) The value line enigma 1965-1978: A case sudy of performance evaluaion issues, Journal of Financial Economics 10, pp. 289 322. Diefenbach, Rober E., (1972) How good is insiuional brokerage research?, Financial Analyss Journal 28, pp. 54 60. Fabrizio S. (2000) Gli «Sudi» prodoi dagli analisi finanziari. Conflii di ineressi, prime evidenze empiriche, Banca Impresa Socieà, n. 2. Grossman S.J., Sigliz J.E. (1980) On he Impossibiliy of informaionally Efficien Markes, The American Economic Review, Vol.70, n.3, pp. 393-408. Groh J., Lewellen W., Schlarbaum G., and Lease R. (1979) An analysis of brokerage house securiies recommendaions, Financial Analyss Journal 35, 32 40. Jegadeesh N., Kim J., Krische S. D., Lee C.M.C. (2004) Analyzing he analyss: When do recommendaions add value?, The Journal of Finance, Vol. 59, n. 3, pp. 1082-1124. Lyon J., Barber B., Tsai C-L (1999) Improved Mehods for Tess of Long-Run Abnormal Sock Reurns, The Journal of Finance, Vol. 59, n.1, pp. 165-201. Sickel S.E. (1995) The anaomy of he performance of buy and sell recommendaions, Financial Analyss Journal, pp. 25-39. Von Nandelsadh A. (2003) Invesmen Informaion in Analyss Recommendaions, in Essay on Financial Analyss Forecass and Recommendaions, Publicaions of he Swedish School of Economics and Business Adminisraion, n.116, pp. 79-105. Womack K.L. (1996) Do Brokerage Analyss' Recommendaions Have Invesmen Value?, The Journal of Finance, Vol. 51, pp. 137-167. 24