Survey							
                            
		                
		                * Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Portfolio Monitoring* Richard Michaud, David Esch, Robert Michaud New Frontier Advisors Boston, MA 02110 Presented to: QWAFAFEW NYC September 27, 2012 * Forthcoming: Michaud, Esch, Michaud, 2012. “Portfolio Monitoring in Theory and Practice,” Journal Of Investment Management. © 2007 Richard Michaud and Robert Michaud © 2011 Richard Michaud and Robert Michaud About New Frontier • Institutional research and investment advisory firm • Inventors and authors in investment technology • Michaud and Michaud, Efficient Asset Management, 1998, Harvard, 2008., 2nd Edition, Oxford • NFA is unique: • • • • Institutional investors who use our own software Global software providers who manage money Published authors in books and refereed journals Four U.S. patents, two pending 22 © 2011 New Frontier Management Company, LLC Current Portfolio Monitoring Ad Hoc  Calendar rebalancing  Monthly, quarterly, yearly, three years, every five minutes  Asset weight hurdle ranges  Drifted portfolio relative to neutral or optimal weights  Ranges typically vary based on asset volatilities  No theory to support practice  Not portfolio based rules  Often trading in noise or not trading when useful 3 17 © 2011 New Frontier Management Company, LLC True Portfolio Monitoring  A statistical similarity test:  Is the current drifted or given candidate portfolio statistically similar or different relative to optimal  If statistically similar, don’t trade  If statistically different, trade  Presentation scope:  Decision whether or not to trade  How to trade or how much to trade is a separate issue 4 17 © 2011 New Frontier Management Company, LLC Academic Portfolio Similarity Tests  Shanken (1985), Jobson and Korkie (1985), Levy and Roll (2010)  Tests of CAPM  Is “market” statistically mean-variance (MV) efficient  Limitations of academic tests  Analytical tests assume unconstrained MV optimization  Hotellings T2 and other analytic methods  Not useful for investment practice  Practice requires linear inequality constraints  Constraints part of defining test statistic  See Markowitz (2005) why constraints essential 5 17 © 2011 New Frontier Management Company, LLC First Constrained Portfolio Similarity Test  Michaud (1998, Ch. 7)  Portfolio distance function relative to Michaud frontier  Uses patented resampling technology  Computes need-to-trade probability  Relative to thousands of simulated investment scenarios  Technology used in NFA’s World Gold Council reports 6 17 © 2011 New Frontier Management Company, LLC Resampling and the Michaud Frontier 67 © 2011 New Frontier Management Company, LLC Statistical Portfolio Monitoring Illustrated 88 © 2011 New Frontier Management Company, LLC What the Monitoring Rule Computes  Associated simulated optimal portfolios provides a distance scale for monitoring portfolios  Portfolio distance function (one example)  Relative variance function = (P – P*) (P – P*)  A measure of distance in N-dimensional portfolio space  Sort distance low to high distribution  Defines probability scale from 0 to 99%  Compute distance from current to optimal  Defines probabilistically how far current from optimal 9 17 © 2011 New Frontier Management Company, LLC What the Rule Means  10% need-to-trade probability means  Portfolio distance is 10% as far as others in distribution  75% or more probability may indicate trading is recommendable  50% probability often a useful default value  Balance between avoiding noise trading and being able to detect true deviations from optimality. 10 17 © 2011 New Frontier Management Company, LLC Using Portfolio Monitoring Rule  Decide on level of probability for trading  L = Probability level for trading  Recommend trading if probability > L L depends on many investment and client issues  Investment Styles:  High levels -- value managers?  Lower levels -- growth managers?  Client Preferences, investment horizon  Specialized investment classes  Way to monitor universe of managed accounts  Portfolio monitoring automation 11 12 © 2011 New Frontier Management Company, LLC Limitations of the Original Michaud (1998) Rule 12 © 2011 New Frontier Management Company, LLC Limitations of Michaud (1998) Test  Low statistical power  Infrequently rejects no-need-to-trade null hypothesis  Poor power at high end of frontier 13 17 © 2011 New Frontier Management Company, LLC Meta-Resampling Solution  Patented meta-resampling (Michaud and Michaud 2002, 2008)  Associates resampled with Michaud efficient portfolios  Each simulated “parent” MV efficient frontier spawns a “child” resampled efficient frontier  Associated child resampled efficient frontier portfolios used to compute distance probability  Greatly enhanced statistical power  Nearly uniform power across frontier 14 17 © 2011 New Frontier Management Company, LLC Michaud Frontier Associated Meta-Resampled Portfolios 12 estimated average return (%) 10 8 6 4 2 0 0 5 15 © 2011 New Frontier Management Company, LLC 10 15 standard deviation (%) 20 25 Highly Compute Intensive Process  Use better computer technology  Multi-core computers  Network multi-core  Cloud computing 16 17 © 2011 New Frontier Management Company, LLC Still A Persistent Problem in Practice  Need-to-trade probabilities often seemed too low in actual practice 17 17 © 2011 New Frontier Management Company, LLC The Common Information Issue 18 © 2011 New Frontier Management Company, LLC The Common Information Issue  Information in current portfolio often based on similar information in new optimal  Common information means two portfolios similar all things equal  Need-to-trade probability necessarily small  Test is no-trading-biased in presence of common information  Michaud-Esch-Michaud conditional monitoring rule  A new scale that includes common information  Dramatically enhanced power for many practical applications  Realistically sensitive to changes in current vs. optimal  Three levels of resampling in general case 19 19 © 2011 New Frontier Management Company, LLC Illustrating Conditional Monitoring Algorithm  One year ago optimal portfolio P0  X0= [x1,x2,…,x60] = defines original risk-return distribution  New optimal portfolio P*  Xnew = [x13,x2,…,x72] = defines new risk-return distribution  48 months of common information: [x13,x2,…,x60]  Compute meta-resampled portfolios (simplest case)  Compute k = random draws = 12 from Xnew distribution  Add to common 48 months: [x13,x2,…,x60] = sim distribution  Compute meta-sim optimal and distance to P*  Repeat above many times  Sort and define distance distribution  Compute P0 distance to optimal and percentile in distance distribution (conditional need-to-trade probability C(k)) 20 4 © 2011 New Frontier Management Company, LLC Actual Case: Conditional Monitoring Rule 21 © 2011 New Frontier Management Company, LLC Applications  A measure of regime changes in markets  Assume a long-term strategic optimal portfolio  In drifted period  Minimal market volatility – little need to trade  High market volatility – likely need to trade  Return distribution generalizations  Simulations can be based on any distribution  We generally use t-distribution 22 4 © 2011 New Frontier Management Company, LLC Summary  Portfolio monitoring an essential asset management function  Prior methods ad hoc, academic methods invalid  Patented first practical monitoring rule Michaud (1998)  Limited statistical power  Patented Meta-resampling rule Michaud and Michaud (2002)  Enhanced statistical power across frontier  Customizable to asset management processes  Michaud-Esch-Michaud conditional monitoring algorithm  Common information, increased statistical power  Highly compute intensive procedures  Just finance catching up to real statistics 23 28 © 2011 New Frontier Management Company, LLC Extensions  Potential for large-scale automatable portfolio monitoring  A statistical context for general quadratic programming applications  Process monitoring and multivariate regression in the context of linear constraints and overlapping data 24 4 © 2011 New Frontier Management Company, LLC Thank You New NewFrontier FrontierAdvisors, Advisors,LLC LLC Boston,MA MA 02110 02110 NFABoston, SAA Portfolios www.newfrontieradvisors.com www.newfrontieradvisors.com 25 25 © 2011 New Frontier Management Company, LLC Richard O. Michaud       President, Chief Investment Officer Co-inventor (with Robert Michaud) of Michaud Resampled Efficient Frontier™, three other patents, two pending Author: Efficient Asset Management, 1998. Oxford University Press, 2001, 2nd Edition 2008 (with Robert Michaud) Many academic and practitioner refereed journal articles CFA Institute monograph on global asset management. Prior positions include:  Acadian Asset Management; Merrill Lynch  Graham and Dodd winner for work on optimization  Former Director and research director of the “Q” Group  Advisory Board member, Journal Of Investment Management  Former Editorial Board member Financial Analysts Journal, Journal Of Investment Management 26 © 2011 New Frontier Management Company, LLC