{"id":306,"date":"2004-12-01T09:00:44","date_gmt":"2004-12-01T09:00:44","guid":{"rendered":"http:\/\/s1094408.instanturl.net\/quantitativegroup.com\/?p=306"},"modified":"2016-07-19T12:00:56","modified_gmt":"2016-07-19T16:00:56","slug":"esis","status":"publish","type":"post","link":"http:\/\/www.quantresearchgroup.com\/?p=306","title":{"rendered":"How to Improve Earnings Revision & Surprise"},"content":{"rendered":"

This paper is the first of a series where we will share the results[1]<\/a> of our research using detail analyst forecast data from Thomson Financial to develop new classes of powerful component factors called ESIS (Estimate Specific Information Score). The ESIS platform is used to gauge the relationship of a range of analyst forecasts and their informational value.\u00a0 While future studies will examine ESIS formulations for non-earnings forecast data, the following research reviews the measure\u2019s application and strength in improving existing earnings based models. In this context, ESIS utilizes earnings estimate data from different fiscal periods to generate an enhanced securities ranking system whose objective is to capture the degree to which the U.S. market assimilates (or responds to) information in price formation.\u00a0 Research indicates the market\u2019s response to released earnings information is incomplete (Stickel 1991, 1995; Chan, Jegadeesh, and Lakonishok 1996; Womack 1996.)\u00a0 ESIS serves as a proxy for analyst characteristics that are manifested in the forecasting of company information and the inefficient dissemination thereof.<\/p>\n

Maximize detail analyst forecasts information<\/em><\/h3>\n

ESIS leverages detail analyst data[2]<\/a> to catalog and exploit important characteristics and systematic differences in analysts\u2019 forecasting ability. How does ESIS accomplish this goal?\u00a0 This factor builds on our extensive experience in modeling detail analyst data at DAIS Group, including innovative work on factors that differentiate, among other things, leading and following forecasts (Herzberg, Guo, Brown 1999; Herzberg, Wang 2002).\u00a0 Specifically, QRG considers a leading analyst forecast as one that moves away from or goes through the consensus while a herding or following forecast goes to or toward the consensus.\u00a0 Evidence of herding in earnings forecasts by security analysts is documented through examination of various measures including departure from consensus and timing of forecasts (Hong, Kubik, Solomon 1999)<\/p>\n

To illustrate the general theory, let\u2019s assume three of the many analysts, designated as A1, A2, A3, who cover company XYZ are each forecasting 10 cents for the firm\u2019s next fiscal period. The consensus, including these and other analysts\u2019 estimates, is 15 cents.\u00a0 Let\u2019s further assume that A3 goes to 13 cents or closer to the consensus, A2 goes to 15 cents, and A1 goes to 17 cents or through the consensus.\u00a0 These three estimates probably reflect different assumptions and skill sets. More important, the forecasts are unlikely to contain the same degrees of informational value and reliability.\u00a0 ESIS is designed to capture the information content of analysts\u2019 departure from consensus while accounting for biases and lags.<\/p>\n

Utilizing a proprietary implementation of the concepts for identifying influential analyst characteristics, we compute a score for each security for various forecast periods such as Quarter1, Fiscal Year1, and Fiscal Year2.\u00a0 Each calculated ESIS is aggregated for each security.\u00a0 Using this raw version of the model formulation, we first performed regression analysis to confirm the score\u2019s predictive capability by regressing subsequent returns on the ESIS variable.\u00a0 The results of the analyses showed a positive and meaningful relationship between total returns and ESIS values.<\/p>\n

To build ranking measures for this factor, the raw ESIS values are normalized within each factor\u2019s respective universe.\u00a0 Moreover, typical quality control and data validity are applied to ensure the integrity of the computed raw scores. Lastly, the values for each fiscal period are optimally combined to produce ESIS rankings that quantify the market sentiment that analysts convey to investors in their earnings forecasts.<\/p>\n

ESIS value added implication<\/em><\/h3>\n

How can ESIS add value to existing earnings framework? Although estimate revision, earnings surprise, and similar earnings based models continue to be a mainstay of most managers\u2019 processes, recent analyses indicate these models are not as effective in capturing the variations in return as they did in the 1980\u2019s and 1990\u2019s (Butman 1998;Thomas 2003).\u00a0 While the reasons for the lackluster performance vary, one of the prevailing conclusions is that consensus estimate revisions has lost its status as a proxy for changes in market sentiment (Zeng 2004.)\u00a0 The conclusions on the declining power of revision as a leading market indicator are primarily based on aggregated analyst earnings data and revisions.\u00a0 The use of detail individual analyst forecasts is an important area that most earnings research has heretofore overlooked.\u00a0 The results in the following pages provide some insights on how ESIS and similarly designed platforms can exploit granular analyst information to bolster revisions and other earnings based models.<\/p>\n

Improvement to revision and surprise models<\/em><\/h3>\n

As Figure 1 illustrates, ESIS has moderately high relationships with growth and momentum factors while showing low to negative correlations with value measures. ESIS\u2019 relationship with revision and forecasted surprise (hereafter ETREND and ERF[3]<\/a> are used interchangeably for these two categories of models) could help to enhance strategies that utilize such traditional models.<\/p>\n

\"ESIS2004_Figue1\"<\/a>To determine whether ESIS is suitable in bolstering existing strategies, we conducted numerous single and multifactor back tests for various universes and compared these results with our ETREND and ERF. For the ten-year period starting in January 1993, ESIS posted approximately 900 percent cumulative excess return[1]<\/a> to significantly outpace revision (330 percent) and forecasted surprise (240 percent).\u00a0 The factor\u2019s pattern of superior performance was also evident after the 2000 TMT bubble.<\/p>\n

\"ESIS2004_Figue2\"<\/a>Specifically, ESIS produced an approximate 7 percent average excess return for the TOP 1000 market-cap and more than 16 percent for the S&P Small-cap universe from January 2000 to December 2003.\u00a0 ETREND and ERF results range from approximately negative -1 to negative \u20135 percent during the same period for the large and small market-cap universes, see Table 1.To determine whether ESIS is suitable in bolstering existing \"ESIS2004_Table1-Smaller\"<\/a>strategies, we conducted numerous single and multifactor back tests for various universes and compared these results with our ETREND and ERF. For the ten-year period starting in January 1993, ESIS posted approximately 900 percent cumulative excess return[4]<\/a> to significantly outpace revision (330 percent) and forecasted surprise (240 percent).\u00a0 The factor\u2019s pattern of superior performance was also evident after the 2000 TMT bubble.<\/p>\n

Bolstering factors + strategies<\/em><\/h3>\n

We also evaluated the ESIS ranks\u2019 contribution to strategies that are focused on subsets of the broad market.\u00a0 In sector specific tests that use the MSCI and S&P GICS[5]<\/a> definitions, ESIS displays strong ability to differentiate favorable and unfavorable investments within economic sectors.\u00a0 The factor\u2019s performance relative to ETREND is illustrated in Table 2.\u00a0 As shown, ESIS is a noticeably better indicator of market sentiments in all sectors, except Tele-communication and Utilities.\u00a0 Moreover, a simple multifactor formulation (50\/50 Combo) that equally weighs ESIS and ETREND bested the standalone factor in nine of the ten sectors.<\/p>\n

\"ESIS2004_Table2\"<\/a>For standalone and multifactor applications, ESIS seems to predict total future returns convincingly in various market capitalization universes.\u00a0 As Figure 3 shows, ESIS generated excess return of approximately 17 percent for the TOP 1000 market-cap universe.\u00a0 These results are more than 4 times the performance of the next closest single model, ETREND. In addition, a basic formulation that equally weights ESIS ranks with another earnings factor significantly outperformed in historical back tests. The total annualized excess returns to each ESIS combination model is at least 3 times greater than the standalone revision and forecasted surprise models.<\/p>\n

\"ESIS2004_Figue3\"<\/a>Consistent with our experience and academic research, the above ESIS out performance in large-cap universe suggests applying similar multifactor strategies to a small capitalization universe should produce substantially better results.\u00a0 As Figure 4 confirms, the total annualized excess returns to ESIS and related combinations are 20 percent or more from January 1995[6]<\/a> to December 2003.<\/p>\n

ESIS performance indicates it can serve as a more effective leading indicator of market sentiment than the traditional revision and forecasted surprise models that currently drive many managers\u2019 investment strategy.\u00a0 The use of individual analyst forecasts as the primary input provides QRG the option to address a number of systematic biases as well as differences in ability and competence that are pervasive in the much modeled consensus analyst information.\u00a0 This access to individual analyst views combined with our extensive modeling experience allowed us to quantify much more of the useful earnings estimate information that analysts disseminate.\u00a0 ESIS\u2019 superior results as a component factor coupled with its contribution to multifactor strategies makes it an attractive complement or alternative to revision and forecasted surprise models.<\/p>\n

Research Update: December\u00a02004<\/em><\/p>\n

References:\r\nButman, R. (1998),\u00a0 \u201cEstimate Revision and Earnings Surprise Backtest Review,\u201d 16th<\/sup> Annual Equity Conference, LaQuinta, California.\r\nChan, L. K. N. Jegadeesh, and J. Lakonishok (1996),\u00a0 \u201cMomentum Strategies,\u201d Journal of Finance, <\/em>51, 1681-1713.\r\nClement, M. (1999), \u201cAnalyst Forecast Accuracy: Do Ability, Resources and Portfolio Complexity Matter?\u201d Journal of Accounting and Economics<\/em>, Vol 27, No. 3, 285-303.\r\nJacob, J., T. Lys, and M. Neale (1999),\u00a0 \u201cExpertise in Forecasting Performance of Security Analysts.\u201d\u00a0 Journal of Accounting and Economics, <\/em>Vol. 28, No. 1, 51-82.\r\nHong H. G., J. D. Kubik (2003), \u201cAnalyzing the Analysts: Career Concerns and Biased Earnings Forecast,\u201d Journal of Finance<\/em>, Vol. 58 No. 1.\r\nHong H, J.D. Kubik, and A. Solomon (1999) \u201cSecurity Analyst\u2019s Career Concerns and Herding of Analyst Forecasts,\u201d Working Papers\r\nHerzberg, M. M., and S. Wang (2002), \u201cIdentifying Lead Analysts for Stock Selection,\u201d Journal of Investing<\/em>, 11, 25-35.\r\nHerzberg, M. M., James Guo, and Lawrence Brown (1999), \u201cEnhancing Earnings Predictability Using Individual Analyst Forecasts,\u201d Journal of Investing<\/em>, Vol. 8, No. 2, 15-24.\r\nMikail, M. B., B. R. Walther, and R. H. Willis (1997), \u201cDo Security Analysts Improve Their Performance With Experience?\u201d Journal of Accounting Research, <\/em>Vol. 35, No. 1, 131-157.\r\nStickel, S. E. (1995), \u201cThe Anatomy of the Performance of Buy and Sell Recommendations,\u201d Financial Analysts Journal<\/em>, 51, 25-39.\r\nStickel, S. E. (1991), \u201cCommon Stock Returns Surrounding Earnings Forecast Revisions: More Puzzling Evidence,\u201d The Accounting Review<\/em>, 66, 402-416.\r\nThomas J. W.\u00a0 (2004), \u201cSmart Earnings Forecast on IBES Earnings Data,\u201d Quantitative Research Group Research Update.<\/em> September, 1-5.\r\n___________ (2003),\u00a0 \u201cSmart Earnings Forecast Model Overview,\u201d Quantitative Research Group\u2019s 19th<\/sup> Annual Equity Conference, LaQuinta, California.\r\nWomack, K. L. (1996), \u201cDo Brokerage Analysts\u2019 Recommendations Have Investment Value?\u201d Journal of Finance<\/em>, 51, 137-167.\r\nZeng, Qi (2004), \u201cQuestioning the Significance of Analyst Revision,\u201d Morgan Stanley Equity Research, October 7.<\/pre>\n
[1]<\/a>\u00a0 The hypothetical back tests were performed using data believed to be reliable. Quantitative Research Group (\u201cQRG\u201d) does not guarantee the accuracy or completeness of such information. QRG shall not have any liability or obligation for the information accuracy. In no event will QRG be responsible for special, indirect, incidental, or consequential damages which might be incurred or experienced on account of using or relying on this information.\u00a0 These hypothetical results are not related to or suggestive of any specific use or implementation of the models in managing assets.\u00a0 Past performance is not a guarantee of future performance.
\n[2]<\/a>\u00a0 Source: Thomson Financial under licensing and redistribution agreement.
\n[3]<\/a>\u00a0 ETREND is an earnings revision and ERF is a forecasted surprise model. Contact QRG for details on these models.
\n[4]<\/a>\u00a0 Excess return – defined as the top 20% ranked stocks (Quintile 1) minus the bottom 20% (Quintile 5.)
\n[5]<\/a> GICS – developed and owned by MSCI and S&P.
\n[6]<\/a>\u00a0 January 95 – the inception date of S&P Small-cap benchmark. ESIS historical ranks start in 1989.<\/p>\n","protected":false},"excerpt":{"rendered":"
This paper is the first of a series where we will share the results[1] of our research using detail analyst forecast data from Thomson Financial to develop new classes of powerful component factors called ESIS (Estimate Specific Information Score). The ESIS platform is used to gauge the relationship of a range of analyst forecasts and<\/p>\n
Read More<\/a><\/div>\n","protected":false},"author":2,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[10,9,8],"tags":[],"_links":{"self":[{"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=\/wp\/v2\/posts\/306"}],"collection":[{"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=306"}],"version-history":[{"count":8,"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=\/wp\/v2\/posts\/306\/revisions"}],"predecessor-version":[{"id":811,"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=\/wp\/v2\/posts\/306\/revisions\/811"}],"wp:attachment":[{"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=306"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=306"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/www.quantresearchgroup.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=306"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}