Artur-eea-presentation
0 likes139 views
This document discusses predicting the rankings of financial analysts based on their forecast accuracy and performance. It presents a research design to: 1) Create target rankings of analysts based on past forecast error; 2) Define state variables like market conditions and stock characteristics; 3) Use these variables to predict analyst rankings using a naive Bayes label ranking algorithm; and 4) Evaluate the predictive accuracy of the model versus baseline methods. The results show the model can predict rankings better than baselines in most sectors, with the consensus forecast being the most predictive variable.
Artur-eea-presentation
- 1. Understanding rankings of financial analysts Artur Aiguzhinov1,2 Ana Paula Serra1 Carlos Soares 2 1 CEFUP & Department of Economics, University of Porto, Portugal 2 LIAAD-INESC Porto LA & Department of Economics, University of Porto, Portugal February 25th, 2011 Agent-based computational economics: Computational Finance Eastern Economics Association Conference, New York 1 of 24
- 2. Motivation (1): the value of the recommendations Efficient Market Hypothesis (Fama, 1970); Information gathering costly ⇒ providing possibilities for abnormal returns (Grossman and Stiglitz, 1980; Fama, 1991); On average, recommendations bring value to investors and financial analysts’ accuracy in forecasts is valuable (Womack, 1996; Barber et al., 2001); 2 of 24
- 3. Motivation(2): rankings of analysts StarMine R issues annual rankings of financial analysts: Ranks the analysts based on recommendation performance For each analyst a portfolio is constructed. For each “Buy”/“Sell” recommendation the portfolio is one (two) unit(s) long/short and simultaneously one (two) unit(s) short/long the benchmark. For “Hold” recommendations, the portfolio invests one unit in the benchmark EPS forecast accuracy Single stock Estimating Score (SES): relative accuracy of each analyst’s earnings forecast when compared with their peer 3 of 24
- 4. Issue: Prediction of rankings of analysts Foreknowledge of analyst forecast accuracy is valuable (Brown and Mohammad, 2003) Is it possible to predict these rankings? (StarMine rankings are ex-post) If yes, can we use those predictions into profitable strategies?; Why not predict stock prices instead? Analysts’ relative performance (rankings) rather more predictable than the stock prices Main goal Accurately predict the rankings of financial analysts 4 of 24
- 5. Research contributions Interdisciplinary approach (label ranking algorithm) First paper to identify the variables that discriminate the rankings of analysts analysis of the financial analysts based on state variables concerning market conditions and stock characteristics (instead of analyst characteristics) First paper to predict the rankings of analysts 5 of 24
- 6. Research design: an overview 1. Create (ex-post) rankings of the analysts (target rankings): to establish the rankings we follow models of Clement (1999); Brown (2001); Creamer and Stolfo (2009); 2. Define state variables 3. Identify the most discriminative state variables 4. Predict rankings of the analysts (naive Bayes for label ranking) and evaluate the ranking accuracy 6 of 24
- 7. Database to use ThomsonOne I/B/E/S Detailed History: Quarterly EPS forecasts (1989Q1-2009Q4); ThomsonOne DataStream: Accounting data; Market data; Filtered out stocks less than 3 analysts less than 8 quarters in forecast history 7 of 24
- 8. Description of the data Table: Summary of the data Sector # stocks # brokers # forecasts # forecasts stocks/broker brokers/stock Energy 170 205 3.09 2.75 Industrials 298 320 2.42 2.23 IT 442 413 2.71 2.41 Materials 94 229 2.56 2.43 Total 1004 1167 2.70 2.46 8 of 24
- 9. Average forecasts Average forcasts per broker Quarters Mu 23456 1989 1 1995 1 2000 1 2005 1 2009 4 Sectors: Energy IT Industrials Materials 9 of 24
- 10. Additional data analysis Figure: Average distribution of brokers based on issued forecasts for all stocks 1 2 3 4 5 0.10.20.30.4 Forecasts per quarter %ofanalysts Sectors: Energy Industrials IT Materials 10 of 24
- 11. topN-lastN analysis Table: Average topN-lastN changes of analysts per quarter, N=3 topt lastt topt−1 lastt−1 topt−1 lastt−1 Energy 3.01 2.35 2.34 2.75 Industrials 3.27 2.44 2.43 2.69 IT 3.61 2.64 2.57 3.22 Materials 2.53 2.04 1.90 2.13 11 of 24
- 12. Creating target rankings: indexing of the analysts Based on previous research, we use EPS mean adjusted forecast error (MAFE1 ) as a measure of analysts predicting accuracy: FEq,a,s = |ActEPSq,s − EPSq,a,s| (1) FEq,s = 1 n n a=1 FEq,a,s (2) MAFEq,a,s = FEq,a,s FEq,s (3) 1We scaled the error measure by adding 1 so that the most accurate analyst will receive a higher rank12 of 24
- 13. Define state variables Variables capture market conditions and stock characteristics (Jegadeesh et al., 2004; Creamer and Stolfo, 2009): Market Conditions: Market volatility (MKT) Consensus Analyst Variables Lagged (consensus) forecasting error (FELAG) Change in consensus (consensus) Earnings Momentum: Standardized Unexpected Earnings (SUE) Growth Indicators: Sales growth (SG) Accounting Fundamentals: Total accruals to total assets ratio (TA) Valuation multiples: Earnings-to-price ratio (EP) 13 of 24
- 14. Define state variables Naive Bayes algorithm requires that continuous variables be converted into discrete discretization using the size bins method (Dougherty et al., 1995) 4 bins High Medium High Medium Low Low 14 of 24
- 15. Discriminative Value Лист1 First Step: Calculate the rankings similarity matrix Quarters Ind. Variables Rankings Similarity matrix X1 X2 X3 X4 a b c 1 2 3 4 5 1 c1 d2 d3 c4 1 2 3 1 1,00 0,25 1,00 0,00 0,00 2 d1 c2 a3 b4 2 3 1 2 0,25 1,00 0,25 0,75 0,75 3 a1 b2 d3 a4 1 2 3 3 1,00 0,25 1,00 0,00 0,00 4 b1 d2 d3 c4 3 2 1 4 0,00 0,75 0,00 1,00 1,00 5 c1 a2 d3 d4 3 2 1 5 0,00 0,75 0,00 1,00 1,00 15 of 24
- 16. Discriminative Value Bins average similarity Weights Weighted average a1 vs. b1 0.00 1 0.00 a1 vs. c1 0.50 2 1.00 a1 vs. d1 0.25 3 0.75 b1 vs. c1 0.50 1 0.50 b1 vs. d1 0.75 2 1.50 c1 vs. d1 0.50 1 0.50 0.708 Discriminative Power : 1-0.708=0.292 The higher the discriminative power, the more different are the rankings between one and the other state of the world 16 of 24
- 17. Results: Discriminative power Table: Discriminative power of independent variables Sectors FELAG SUE consensus EP SG TA MKT Energy 0.163 0.194 0.224 0.218 0.183 0.186 0.222 IT 0.207 0.214 0.220 0.196 0.183 0.197 0.194 Industrials 0.185 0.179 0.192 0.210 0.200 0.224 0.190 Materials 0.199 0.189 0.258 0.218 0.203 0.220 0.183 17 of 24
- 18. Predict with Label Ranking Algorithm Naive Bayes algorithm for label ranking (Aiguzhinov et al., 2010): non parametric technique that relies on the similarities of the rankings predicts rankings conditional on the values of the state variables Alternative baseline ranking methods: default (ranking based on the average rank of each label) naive (previous quarter ranking) Accuracy of the methods: Spearman’s rank correlation Figure: Time line of the predicted ˆπ and the target π rankings 18 of 24
- 19. Results: Label rankings Table: Ranking accuracy of the naive Bayes ranking method and the two baselines. NBr default naive ranking Sector mean std.dev mean std.dev mean std.dev Energy 0.037 0.060 0.060 0.087 0.064 0.060 IT -0.005 0.045 0.088 0.075 0.043 0.036 Industrials 0.010 0.050 0.078 0.074 0.033 0.046 Materials 0.012 0.067 0.076 0.073 0.041 0.056 19 of 24
- 20. Differences(1) Figure: Differences in ranking accuracy of naive Bayes and the default rankings 0 50 100 150 1.00.50.00.51.0 Energy Stocks DifferencebetweenmeanaccuracyofNbranddefaultranking 0 50 100 150 200 250 300 1.00.50.00.51.0 Industrials Stocks DifferencebetweenmeanaccuracyofNbranddefaultranking 0 100 200 300 400 1.00.50.00.51.0 IT Stocks DifferencebetweenmeanaccuracyofNbranddefaultranking 0 20 40 60 80 1.00.50.00.51.0 Materials Stocks DifferencebetweenmeanaccuracyofNbranddefaultranking 20 of 24
- 21. Differeces(2) Figure: Differences in ranking accuracy of naive Bayes and the naive rankings 0 50 100 150 1.00.50.00.51.0 Energy Stocks DifferencebetweenmeanaccuracyofNbrandnaiveranking 0 50 100 150 200 250 300 1.00.50.00.51.0 Industrials Stocks DifferencebetweenmeanaccuracyofNbrandnaiveranking 0 100 200 300 400 1.00.50.00.51.0 IT Stocks DifferencebetweenmeanaccuracyofNbrandnaiveranking 0 20 40 60 80 1.00.50.00.51.0 Materials Stocks DifferencebetweenmeanaccuracyofNbrandnaiveranking 21 of 24
- 22. Conclusions Discriminative power analysis identifies Consensus as the most discriminative variable in most of the sectors There is a room for improving of label ranking algorithm in particular refining predictor state variables 22 of 24
- 23. References (1) Aiguzhinov, Artur, Carlos Soares, and Ana Serra (2010), “A similarity-based adaptation of naive bayes for label ranking: Application to the metalearning problem of algorithm recommendation.” In Discovery Science (Bernhard Pfahringer, Geoff Holmes, and Achim Hoffmann, eds.), volume 6332 of Lecture Notes in Computer Science, 16–26, Springer Berlin, Heidelberg. Barber, B., R. Lehavy, M. McNichols, and B. Trueman (2001), “Can Investors Profit from the Prophets? Security Analyst Recommendations and Stock Returns.” The Journal of Finance, 56, 531–563. Brown, L. (2001), “How Important is Past Analyst Earnings Forecast Accuracy?” Financial Analysts Journal, 57, 44–49. Brown, L.D. and E. Mohammad (2003), “The Predictive Value of Analyst Characteristics.” Journal of Accounting, Auditing and Finance, 18. 23 of 24
- 24. References (2) Clement, M.B. (1999), “Analyst forecast accuracy: Do ability, resources, and portfolio complexity matter?” Journal of Accounting and Economics, 27, 285–303. Creamer, G. and S. Stolfo (2009), “A link mining algorithm for earnings forecast and trading.” Data Mining and Knowledge Discovery, 18, 419–445. Dougherty, J., R. Kohavi, and M. Sahami (1995), “Supervised and unsupervised discretization of continuous features.” In MACHINE LEARNING-INTERNATIONAL WORKSHOP, 194–202, MORGAN KAUFMANN PUBLISHERS, INC. Fama, E.F. (1970), “Efficient Capital Markets: A Review of Empirical Work.” The Journal of Finance, 25, 383–417. Fama, E.F. (1991), “Efficient Capital Markets: II.” The Journal of Finance, 46, 1575–1617. 24 of 24
- 25. References (3) Grossman, S.J. and J.E. Stiglitz (1980), “On the Impossibility of Informationally Efficient Prices.” American Economic Review, 70, 393–408. Jegadeesh, N., J. Kim, S.D. Krische, and C.M.C. Lee (2004), “Analyzing the Analysts: When Do Recommendations Add Value?” The Journal of Finance, 59, 1083–1124. Vogt, M., JW Godden, and J. Bajorath (2007), “Bayesian interpretation of a distance function for navigating high-dimensional descriptor spaces.” Journal of chemical information and modeling, 47, 39–46. Womack, K.L. (1996), “Do Brokerage Analysts’ Recommendations Have Investment Value?” The Journal of Finance, 51, 137–168. 25 of 24
- 26. Similarity-based Naive Bayes for Label Ranking: Prior probability of label ranking Table: Demonstration of the prior probability for label ranking Quarters x1 x2 x3 x4 Ranks Alex Brown Craig 1 High Low High Medium 1 2 3 2 High High High Low 2 3 1 3 Medium Medium High Low 1 2 3 4 Low Low Low High 1 3 2 ... ... ... ... ... ... ... ... 14 Medium High High Medium 1 2 3 15 High Medium High Low 3 1 2 Maximizing the likelihood is equivalent to minimizing the distance (i.e., maximizing the similarity) in a Euclidean space Vogt et al. (2007)
- 27. Label ranking: formalization Instance: X ⊆ {V1, . . . , Vm} Labels: L = {λ1, . . . , λk } Output: Y = ΠL Training set: T = {xi , yi }i∈{1,...,n} ⊆ X × Y Learn a mapping h : X → Y such that a loss function is minimized: = n i=1 ρ(πi , ˆπi ) n (4) with ρ being a Spearman correlation coefficient: ρ(π, ˆπ) = 1 − 6 k j=1(πj − ˆπj )2 k3 − k (5) where π and ˆπ are, respectively, the target and predicted rankings for a given instance.
- 28. Posterior probability of label ranking Proir probability of label ranking: PLR (π) = n i=1 ρ(π, πi ) n (6) Conditional probability of label ranking: PLR (va,i |π) = i:xi,a=va,i ρ(π, πi ) |{i : xi,a = va,i }| (7) Estimated ranking: ˆπ = arg max π∈ΠL PLR (π) m a=1 PLR (xi,a|π) (8)