fma.ny.presentation
0 likes219 views
This document discusses research on predicting the rankings of financial analysts. The goals are to accurately predict analyst rankings and identify variables that can discriminate between rankings. The research uses historical quarterly earnings per share forecasts and market/accounting data to initially rank analysts based on forecast accuracy. State variables are then used to predict these rankings using a Naive Bayes label ranking algorithm. Results show the predictions outperform default and naive rankings for many stocks, and certain variables like total accruals have higher discriminative power between rankings. The research aims to contribute new methods for analyzing and predicting financial analyst performance rankings.
fma.ny.presentation
- 1. Understanding rankings of financial analysts Artur Aiguzhinov1 Carlos Soares1 Ana Paula Serra2 1 LIAAD-INESC Porto LA & Faculdade de Economia da Universidade do Porto 2 Faculdade de Economia da Universidade do Porto & CEFUP - Centro de Economia e Finan¸cas da Universidade do Porto October 23rd, 2010 FMA Annual Meeting, New York 1 of 23
- 2. Motivation: the value of the recommendations Efficient Market Hypothesis (Fama, 1970); High information costs provide possibilities for abnormal returns (Grossman & Stiglitz, 1980); On average, recommendations bring value to investors (Womack, 1996); Analysts’ accuracy in forecasts is valuable (Brown & Mohammad, 2003); 2 of 23
- 3. Motivation: rankings of the analysts StarMine R issues annual analyst rankings: Ranks the analysts based on recommendation performance and EPS forecast accuracy; Why not to predict stock prices directly? Analysts’ relative performance (rankings) is more predictable than the stock prices. Is it possible to predict these rankings?: If yes, can we use those predictions in profitable strategy?; 3 of 23
- 4. The Goals of the Research Accurately predict the rankings of financial analysts; 4 of 23
- 5. Research contributions Interdisciplinary approach to an interesting research topic; Financial Economics contributions: analysis of the financial analysts based on state variables concerning market conditions and stock characteristics; first methodology to predict the rankings; verify if there is a ranking consistency over time; identify variables that discriminate the rankings; 5 of 23
- 6. Research design: an overview Initial rankings of the analysts (target rankings): Analysts evaluation models (Clement, 1999; Brown, 2001; Creamer & Stolfo, 2009); Predict rankings of the analysts (label ranking): Evaluate the ranking accuracy; Identify discriminative independent variables; 6 of 23
- 7. Database to use ThomsonOne I/B/E/S Detailed History: Quarterly EPS forecasts (1989Q1-2009Q4); ThomsonOne DataStream: Accounting data; Market data; 7 of 23
- 8. Description of the data Table: Summary of the data Sector # analysts # stocks mean forecasts mean stocks per stock per quarter per analysts per quarter Energy 135 34 7.57 1.91 Industrials 208 66 3.66 1.05 Materials 147 30 5.14 1.16 IT 301 106 7.85 2.76 Total 791 236 6.06 1.72 8 of 23
- 9. Average forecasts Figure: Average issued forecasts per quarter 0 20 40 60 80 010203040506070 Quarters Averagenumberofforecastsperquarter Sectors Energy Industrials IT Materials 9 of 23
- 10. Creating target rankings: indexing of the analysts Based on previous research, we use EPS mean adjusted forecast error (MAFE) 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) 10 of 23
- 11. Context characterization: state variables Variables that describe market conditions and stock characteristics (Jegadeesh, Kim, Krische, & Lee, 2004): Analysts variables: Lagged forecasting error; Change in consensus; Earnings momentum: Standardized Unexpected Earnings; Growth indicators: Sales growth; Fundamentals: Total accruals to total assets ratio; Valuation multiples: Earnings-to-price ratio; Market volatility; 11 of 23
- 12. Label ranking algorithm Table: Example of analysts rankings based on the observed variables x1 . . . x4 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 5 Medium High High Medium 1 2 3 6 High Medium High Low 3 1 2 Naive Bayes algorithm for label ranking (Aiguzhinov, Soares, & Serra, 2010): ˆπ = arg max π∈ΠL PLR (π) m a=1 PLR (xi,a|π) (4) 12 of 23
- 13. Predicted vs. default ranking Table: Summary of the results compared to default ranking Sectors Outperformed Outperformed- # stocks with p-values the default ranking to-total rate 1% 5% 10% Energy 18 0.53 7 9 9 Industrials 31 0.47 16 18 23 Materials 12 0.40 5 7 8 IT 51 0.48 18 27 30 Total 112 0.47 46 61 70 13 of 23
- 14. Predicted vs naive ranking Table: Summary of the results compared to naive ranking Sectors Outperformed Outperformed- # stocks with p-values the naive ranking to-total rate 1% 5% 10% Energy 11 0.32 5 7 7 Industrials 27 0.41 5 6 7 Materials 11 0.37 0 0 1 IT 36 0.34 1 1 1 Total 85 0.36 11 14 16 14 of 23
- 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 23
- 16. Discriminative Value Лист1 Second step: X1=(a1,b1,c1,d1) a1 vs b1 X1 1 2 3 4 5 1 c1 1 2 3 1 1,00 0,25 1,00 0,00 0,00 2 d1 2 3 1 2 0,25 1,00 0,25 0,75 0,75 3 a1 1 2 3 3 1,00 0,25 1,00 0,00 0,00 4 b1 3 2 1 4 0,00 0,75 0,00 1,00 1,00 5 c1 3 2 1 5 0,00 0,75 0,00 1,00 1,00 mean(a1 vs b1) = (0)/1=0 16 of 23
- 17. Discriminative Value Лист1 a1 vs c1 X1 1 2 3 4 5 1 c1 1 2 3 1 1,00 0,25 1,00 0,00 0,00 2 d1 2 3 1 2 0,25 1,00 0,25 0,75 0,75 3 a1 1 2 3 3 1,00 0,25 1,00 0,00 0,00 4 b1 3 2 1 4 0,00 0,75 0,00 1,00 1,00 5 c1 3 2 1 5 0,00 0,75 0,00 1,00 1,00 mean(a1 vs c1) = (1+0)/2=0,5 17 of 23
- 18. Discriminative Value Лист1 a1 vs d1 1 2 3 4 5 1 1,00 0,25 1,00 0,00 0,00 2 0,25 1,00 0,25 0,75 0,75 3 1,00 0,25 1,00 0,00 0,00 4 0,00 0,75 0,00 1,00 1,00 5 0,00 0,75 0,00 1,00 1,00 mean(a1 vs d1) = (0,25)/1=0,25 18 of 23
- 19. Discriminative Value X1 average 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 19 of 23
- 20. Discriminative power Table: Discriminative power of independent variables Sectors FELAG SUE consensus EP SG TA MKT Energy 0.235 0.211 0.234 0.208 0.194 0.341 0.148 Industrials 0.248 0.230 0.279 0.233 0.263 0.282 0.089 Materials 0.180 0.197 0.278 0.232 0.156 0.378 0.051 IT 0.213 0.238 0.243 0.230 0.185 0.298 0.155 Total 0.219 0.219 0.258 0.226 0.199 0.325 0.111 20 of 23
- 21. References (1) Aiguzhinov, A., Soares, C., & Serra, A. P. (2010). A similarity-based adaptation of naive bayes for label ranking: Application to the metalearning problem of algorithm recommendation. In B. Pfahringer, G. Holmes, & A. Hoffmann (Eds.), Discovery science (Vol. 6332, pp. 16–26). Springer. Black, F., & Litterman, R. (1992). Global portfolio optimization. Financial Analysts Journal, 48(5), 28–43. Brazdil, P., Soares, C., & Costa, J. (2003). Ranking Learning Algorithms: Using IBL and Meta-Learning on Accuracy and Time Results. Machine Learning, 50(3), 251–277. Brown, L. (2001). How Important is Past Analyst Earnings Forecast Accuracy? Financial Analysts Journal, 57(6), 44–49. Brown, L., & Mohammad, E. (2003). The Predictive Value of Analyst Characteristics. Journal of Accounting, Auditing and Finance, 18(4). 21 of 23
- 22. References (2) Clement, M. (1999). Analyst forecast accuracy: Do ability, resources, and portfolio complexity matter? Journal of Accounting and Economics, 27(3), 285–303. Creamer, G., & Stolfo, S. (2009). A link mining algorithm for earnings forecast and trading. Data Mining and Knowledge Discovery, 18(3), 419–445. Fama, E. (1970). Efficient Capital Markets: A Review of Empirical Work. The Journal of Finance, 25, 383–417. Grauer, R. (2008). Benchmarking measures of investment performance with perfect-foresight and bankrupt asset allocation strategies. The Journal of Portfolio Management, 34(4), 43–57. Grossman, S., & Stiglitz, J. (1980). On the Impossibility of Informationally Efficient Prices. American Economic Review, 70, 393–408. H¨ullermeier, E., F¨urnkranz, J., Cheng, W., & Brinker, K. (2008). Label ranking by learning pairwise preferences. Artificial Intelligence, 172(2008), 1897–1916. 22 of 23
- 23. References (3) Jegadeesh, N., Kim, J., Krische, S., & Lee, C. (2004). Analyzing the Analysts: When Do Recommendations Add Value? The Journal of Finance, 59(3), 1083–1124. Ljungqvist, A., Malloy, C., & Marston, F. (2009). Rewriting history. The Journal of Finance, 64(4), 1935–1960. Vembu, S., & G¨artner, T. (2010, October). Preference learning. Springer. Vogt, M., Godden, J., & Bajorath, J. (2007). Bayesian interpretation of a distance function for navigating high-dimensional descriptor spaces. Journal of chemical information and modeling, 47(1), 39-46. Womack, K. (1996). Do Brokerage Analysts’ Recommendations Have Investment Value? The Journal of Finance, 51, 137–168. 23 of 23
- 24. 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, Godden, & Bajorath, 2007)
- 25. 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 (5) with ρ being a Spearman correlation coefficient: ρ(π, ˆπ) = 1 − 6 k j=1(πj − ˆπj )2 k3 − k (6) where π and ˆπ are, respectively, the target and predicted rankings for a given instance.
- 26. Posterior probability of label ranking Proir probability of label ranking: PLR (π) = n i=1 ρ(π, πi ) n (7) Conditional probability of label ranking: PLR (va,i |π) = i:xi,a=va,i ρ(π, πi ) |{i : xi,a = va,i }| (8) Estimated ranking: ˆπ = arg max π∈ΠL PLR (π) m a=1 PLR (xi,a|π) (9)