Sector Investing. Prepared for AAII Silicon Valley Computerized Investing Subgroup Michael J Begley March 6, 2007

Transcription

1 Sector Investing Prepared for AAII Silicon Valley Computerized Investing Subgroup Michael J Begley March 6, 2007

2 Sector Classifications Industry Style Value Growth Market Cap Small Medium Large Country 03/06/2007 2

3 Approaches to Sector Analysis Technical Analysis Independent Sector Analysis Prior AAII talk by Bob Feretich Provides good overview of sector investing Buy out of favor low relative volatility sectors Bob reports excellent results See Appendix for how to access his presentation Rotation Relative Strength Analysis Positive Momentum (buy strength) Negative Momentum (buy dips) Fundamental Analysis Business cycle analysis Currency analysis for Country Funds Fundamental Model for each sector Valuation, Earnings Growth, etc Primary industry variables interest rates, oil prices, etc 03/06/2007 3

4 Sector Investing Research 1. Financial Analyst Journal (FAJ) 2. Investment Books 3. Other Sector Investors

5 Book: Opportunity Investing by Gerald Appel I recommend this book. There is a lot here to improve your portfolio. Appel s approach to Mutual Funds Use rotation strategies based on relative strength Use funds with Best long, intermediate, and short term records Low fees Very Important -- the lowest volatility Ranking periods used 26 weeks 13 weeks 52 weeks 13weeks + 52 weeks Hold periods range from 13 weeks to 52 weeks 03/06/2007 5

6 NoLoad FundX Investment Newsletter ranks Mutual Funds in several volatility classes Ranking is a compound ranking Create a fund score by adding its 4 week return, 13 week return, 26 week return, and 52 week return. Rank all mutual funds in each volatility class by this score and invest in the top ranked funds. Hulbert Financial Digest performance ranking for this simple approach: 03/06/2007 6

7 FAJ1 Sector Effects in Developed vs. Emerging Markets Relative importance of industrial factors vs. country factors in global equity returns Industry factor is what industry a stock is in independent of its country location; Country factor is what country a stock is located in independent of its industry. Recent years have seen a change from a strict dominance of country factors Developed Countries industrial factors are now as important as country factors Emerging Markets country factors dominate Authors (ref 1) recommend a 2 factor model Take-away for single factor models Invest by industry across developed countries Consider investing by country for both developed and emerging markets 03/06/2007 7

8 FAJ2 Understanding Momentum Paper decomposes momentum returns into: Sector rotation effects Country rotation effects Company specific return effects Different factors drive momentum in small and large cap companies Industry effects dominate for large cap Company factors dominate for small cap 03/06/2007 8

9 FAJ3: Style Management in Equity Country Allocation Investigated returns 1/75 through 8/03 across 18 MSCI country indexes and found Relative strength (momentum) and relative value (B/P) each produced excess returns at the country index level Returns of the two styles were negatively correlated Strategies for allocating capital between the two styles were investigated: Fixed styles either relative value or relative strength Style Diversification 50:50 allocation between styles Style Timing 100% switch between the two based on the 12 month return of the global market relative to its historical return If above choose relative strength If below choose relative value Style Management Combo of 50% Style Diversification and 50% Style Switching Geometric monthly excess (top 25% less bottom 25%) returns found: Style Management: 1.24% (significant at 5% level) Style Timing: 1.17% (significant at 5% level) Style Diversification: 0.80% (significant at 5% level) Relative Strength: 0.65% (significant at 5% level) Relative Value: 0.22% Future Study Very interesting if we can obtain the P/B time series for county indexes. Any sources for this??? Can at least develop timing strategy for when to do sector rotation and when to go elsewhere 03/06/2007 9

10 FAJ5: Industry Momentum and Sector Mutual Funds Investigates momentum with Fidelity Select Funds from May 1989 to April 1999 Selected findings for his top portfolios Rank period, hold period, # funds held, CRR, SD, Sharpe 6,12,3 20.5% 25.5% ,12,6 22.7% 22.1% ,12,3 26.5% 24.6% ,12,6 22.1% 22.1% 0.76 SP % 15.7% 0.86 Top 10% Domestic Eq MFs 20.3% 20.2% 0.66 Portfolio Measurements Betas were greater than 1 Alphas ranged from.4% to 5.4% (but not found to be significant!) Regressed excess return of high portfolio against several economic variables Found a significant (1% level) positive relationship with the high yield bond return premium over the T-bond return At the 5% significant level one portfolio showed a positive return to the T-bond return premium over the T-bill return premium Comments Long hold times helped in bull market Holding 3 funds gave him higher volatility than the market Sharpe ratios did not match SP500 but neither did top 10% domestic mutual funds Economic variable relationships do not bode well for this strategy today 03/06/

11 Proposed Sector Rotation Model

12 Proposed Sector Rotation Model - Goals Try to replicate the results of journals, NoLoad FundX, Appel, and others. Explore an Industry Sector Rotation model All calculations and results will be explained Consider all GICS industry sectors plus a few more. Avoid using hindsight to cherry pick the sectors in the investing universe Use the deepest historical price history available (to me) Use a total return data series that adjusts for any splits, dividends and other distributions Minimize survivorship bias in funds used in back test Measure the results and test their significance Then after efficacy is thoroughly demonstrated, apply the model to other sector classifications: country funds, market capitalization and style, and other models. 03/06/

13 Proposed Sector Rotation Method Ranking Period Form Portfolio 1 Holding Period 2 Ranking Period Holding Period Ranking Period 3 Holding Period Rank our n sector funds with a % Price Change score over a trailing ranking period. Purchase a portfolio of the m top ranked funds. Hold portfolio for holding period. Re-rank and repeat the process. 03/06/

14 Proposed Sector Rotation Method Multi-period Scoring and Ranking For each fund in our purchase universe compute % Price Change for each rank period. For each fund sum the % Price Changes across all the rank periods after factoring in the weights for each rank period to create a total score. Rank these scores to compute each fund s final rank. Example below ranks 2 funds using 3 ranking periods. Note, that I am very close to the NoLoad FundX scoring method. 03/06/

15 Simulation Model Technology Excel this is what I used OK for small simple models I am way past this and I don t recommend following my foot steps Amibroker this is what I recommend Code your own rotation algorithms in a simple scripting language Very low cost Data Reads FastTrack data base (and others) directly Downloads free Yahoo data into its own data base 03/06/

16 Data Sources What do we test with? All data for this presentation is from FastTrack data base Price series are adjusted for dividends and all other distributions, splits, etc Creates a total return series This is a requirement as we will rank by historical performance! Drawback subscription service Yahoo Finance Historical data has a field where data is adjusted for splits and distributions I checked a few mutual funds with frequent distributions and these looked OK Drawback does not catch all the distributions (consensus of Feb 2007 Amibroker / FastTrack conference attendees) Suggestion start with Yahoo Finance but move to FastTrack if you get serious about mutual fund modeling. 03/06/

17 Data Considerations Begin with Industry Sectors Requirements Data as deep as possible No (or minimal) survivorship bias Include dividend returns Fidelity Select Funds Data from 9/1/88 in FastTrack data base Total return series includes all distributions Combine funds into the current Global Industry Classification Standard (GICS ), developed by Standard & Poor's (S&P) and Morgan Stanley Capital International (MSCI ). This is the new world order Fidelity reorganized their funds around this schema 10/1/06 ETF s organized around this schema I combine Select Funds in an equal weight manner into GICS sectors Add additional sectors International Funds Real Estate Move on to country funds, market cap and style segments, industry ETFs, etc. 03/06/

18 Fidelity Select mapped to GICS 10/06 Fidelity Reorg narrowed focus of some funds I dropped these. Last 3 sectors are not GICS added to increase number of stock sectors. 03/06/

19 Measurements

20 Measurement Definitions 03/06/

21 Measurement Examples Equity Curve Drawdown 1 (MaxDD) Drawdown 2 Ulcer Index (UI) is the RMS of the data points in the drawdown curve. Draw-Down Curve 03/06/

22 Proposed Rotation Model

23 Model 1 Base Model Parameters A funds Rank Score is the sum of the following: 0.25 x return of the prior 13 weeks 0.50 x return of the prior 26 weeks 0.25 x return of the prior 52 weeks Holding period is 17 weeks Of 13 sectors, invest 2 units of money in the top 3 ranked sectors 1 unit of money in the next 6 ranked sectors No money in the lowest 4 ranked sectors 03/06/

24 Model 1 Base Model BB UP, Down refer to Bob Brinker s Market Timing Calls 03/06/

25 Model 1 and Benchmark Annual Returns 5 Years of Good Performance but less than Benchmark 03/06/

26 Model 1 Base Model Bull Market 1 Irrationally Exuberant Bull Market Bear Market Bull Market 2 03/06/

27 Model 1 Draw-down Analysis 03/06/

28 Result Reporting Table I allow for scoring a fund using a compound rank which adds together the percentage return of the fund over 4 different time periods: 4 weeks, 13 wks, 26 wks, 52 wks. The return over each period is weighted prior to adding them up for a final score. The weights are given here with the first digit the weight for 13 weeks, the second for 26 weeks, etc. There are 13 sector funds in the simulation. In Sector Weights by Rank, the leftmost digit represents the highest rank fund, the second digit represents the 2 nd highest fund, etc. The number of digits in the string equals the number of funds. The number of units of weight allocated to a fund is indicated by the value of the digit which could be 0, 1, or 2. 03/06/

29 Summary Industry Sectors Model 1 Model 1 (base model) beats VFINX on nearly every benchmark Model 2 (base less RE and International) also beats VFINX but not Model 1 I show this for completeness We will continue including RE and International 03/06/

30 Before proceeding further. Are results significant? Make sure returns are significantly different than SP 500 returns Are parameter values robust? Vary parameters and make sure nearby parameters work well also Start Dates Ranking Period Lengths Holding Period Lengths 03/06/

31 Portfolio Performance for Non-Nerds See Computation and Statistical Significance of the alpha and beta estimates of Model 1 slides in Appendix for regression detail. Annual CRR 16.8% Model % Portfolio Alpha VFINX 90 day Treasury (risk free rate) 4.5% Portfolio Beta X axis is Beta a measure of market risk 03/06/

32 Portfolio Performance Comments An alpha of over 7% seems very high to me. Anything that good, there is usually a problem with. Most of the alpha developed during the bear market when the rotation portfolio performance crushed the VFINX performance. If we do not have additional bear markets similar to 2000 to 2003, the rotation portfolio alpha will be much lower. If we have another IE Bull market, I expect the rotation portfolio alpha to be negative as it was during the last IE Bull market There may be other risks that I have not identified that are not measured in this process. My gut tells me the rotation portfolio alpha is positive but is much less than 7%. The measurement of alpha is an area that I continue to investigate. 03/06/

33 Check Alternative Start Dates All start dates have similar profitability with no really bad results. Our start date, 12/29/89, generated better than average results Can get close to average results by dividing money into 4 pools and rotating 1 pool every month. I will go forward with the 12/29/89 start date. 03/06/

34 Check Alternative Ranking Period Lengths All ranking periods are profitable Combined ranks look for more consistent performance. I chose a combination ranking centered around 26 weeks (shaded in green). 03/06/

35 Check Alternative Hold Periods All Holding periods are profitable. Hold Period Length of 17 wk is close to edge of a plateau. For ETF portfolios, consider a 13 wk hold period. I will continue using a 17 wk holding period (which may be used with open ended mutual funds). 03/06/

36 Model 3: Top 3 Sectors Focus all dollars into the top 3 sectors Result: Make more money in bulls and lose more in bears UPI declines a lot. 03/06/

37 Market Timing Bob Brinker (BB) Newsletter makes infrequent market timing calls primarily on fundamental models and recommends simple conservative portfolios of mutual funds. Has made 3 market timing calls in 16+ years Long January 1991 Cash January 2000 Long March 2003 Only 3 calls but he had 4,000 other market days to screw it up with other calls and he did not. Brinker s timing signals are out-of-sample real money signals since January Hulbert s ranking of Bob Brinker s Marketimer newsletter The Red and Green poles on the Equity Curve graphs are Market Timing calls by Bob Brinker. Bottom Line: Get long in a bull, get out of a bear, and don t mess around with minor up and down trends. 03/06/

38 Models 4 to 6: Market Timing Add BB Market Timing Model 4: Base model and go flat during Bear Markets Model 5: Base model and invest in Bond Fund during Bear Markets Model 6: Top 3 Sectors and invest in Bond Fund during Bear Markets Results CRR and UPI are increased 03/06/

39 Model 4 & 5: Base w/ Mkt Timing 03/06/

40 Model 6: Top 3 with Mkt Timing 03/06/

41 Model 7: Base, Mkt Time, 20% Bonds Can mix bonds with stock MF portfolio to decrease draw downs and improve UPI Model 7: Base Model rotation with BB Market Timing Bull Periods: 20% bonds + 80% stock rotation Rebalance back to 20% / 80% every 17 weeks Bear Periods: 100% bonds Good UPI with lower UPI and beats VFINX handily 03/06/

42 Model 7: Base, Mkt Time, 20% Bonds 03/06/

43 Country Fund Models This model follows the journal research which found a significant momentum effect in stocks in a given country. Country Fund ETFs are used to capture this effect. The list of ETFs comes from Timing Cube which has a service to rank these for subscribers. (We use our own rank here.) I used this list except for A few funds not included due to insufficient history or model space Indexes used in place of 2 ETFs due to lack of price history 03/06/

44 Country Fund Models CF3 Top 5 looks most attractive with good performance in Bull2 period CF6 Top 5 with bonds during bear has a good CRR and UPI Country Funds can give you good returns in a bull but try to stand clear in a bear. This model is very risky as it will tend to pick highly correlated funds in one part of the world. At most, a very small allocation of capital is appropriate. In the week ending March 2, 2007, this portfolio lost 8.9%! 03/06/

45 Model CF6: Top5 + BB M.T. + Bonds This graph ends on 3/2/ % DD 03/06/

46 Market Style & Cap Segment Models Market Style and Capitalization Segments Divide market into Market Style Segments Divide market into Market Cap Segments Buy the top ranking funds Parameters the same and we continue with a 13 week hold period. ETF funds used in this model follow. VFINX is the benchmark. 03/06/

47 Market Style & Cap Segment Models Top3 looks good relative to VFINX Bull2 performance lags top Country Fund and Industry Sector Bull2 performance 03/06/

48 Model MSC6: Top3 + MktTime + Bonds 03/06/

49 Model from Gerald Appel Model taken from his book Opportunity Investing 7 Sectors included in model Not as diversified as it could be and no satisfying rational offered for why these sectors He does suggest adding international and income funds to this mix; he also suggests investing in more than 1 fund per sector With exception of Energy, they are low volatility funds which he recommends later in the book. Funds were chosen for their above average historical performance (look ahead bias?) You need to research funds for your own model as some of his 7 funds are closed or have loads which I suggest you avoid. Possible flaws aside, I liked this idea and this model was the inspiration to create my own. 03/06/

50 Appel 7 Model My parameters differed He used 26 weeks while I used my usual compound ranking. He held for 26 weeks and I held for 17 weeks. He actually only rotated 6 of the funds and held Vanguard Health Care continuously; I rotated everything. Appel reports an arithmetic average of 21.2% while I get a CRR of 15.2%. This is still a great showing relative to VFINX. 03/06/

51 Model Appel3: base + BB M.T. + Bonds 03/06/

52 Appel Fund Selection Criteria Low Expense Ratio Low Portfolio Turnover No-Load likely to outperform Load Funds Low Volatility is better Measures with beta and thinks 0.6 to 0.8 is great Higher volatility could work but frequently works against the investor For higher volatility funds he suggests market timing which is beyond the scope of this talk. 03/06/

53 Real Portfolio 11/5/06 Real portfolio started 11/5/06 Inspired by Appel s 7 fund model and his suggestions for choosing funds Portfolio consists of 14 open ended funds Two additional model parameters used Positions are sized by relative fund volatility Fund ranking is adjusted for relative fund volatility During bear markets An additional timing service was employed (Timing Cube - TC) 50% of the portfolio funds were allocated to long/short index positions per TC state So far, model performance has been consistent with the back test simulation 03/06/

54 Real Portfolio 11/5/06 Equal weight model adds value by itself Base Model does well over an 8 year simulation Things go better with Market Timing that works 03/06/

55 Real Portfolio 11/5/06 Bear Market 50% Bonds 50% Long/Short indexes with TC Portfolio started and becomes real This Graph goes to 3/2/07 For Wk Ending 3/2/07: Portfolio Lost 3.35% VFINX lost 4.36% 03/06/

56 Summary Reviewed Sector Momentum research Constructed sector total return data series free of survivor bias and selection bias Demonstrated an industry sector rotation model with some promise Applied this model to other sector classifications There are plenty of starting places to jump off into your own research! 03/06/

57 Appendix

58 Prior AAII Talk by Bob Feretich Bob gave a talk Sector Investing (The Basics) in the Fall of 2005 Bob created a Yahoo Group aaii-sector-investing Go to this url and request a membership invitation Bob s presentation is in the files section of this site In the messages section, Bob gives more detail on calculating the necessary relative volatility z-scores There are some ideas Bob developed that could be carried over into the sector rotation models I presented I have used several relative volatility measures with some success in Scoring and Ranking Funds Position Sizing These ideas were mentioned only in passing in my presentation today I hope to look at these ideas further after I port my models over to Amibroker 03/06/

59 Resources 1 Data Yahoo Finance: Investors FastTrack: Local FastTrack user group Covers FastTrack, Amibroker, and related trading products Leader: Sidney Kaiser, s9kaiser@pacbell.net Send Sid an asking to be put on his meeting notice list. Software for designing and simulating models including rotational models Amibroker: 03/06/

60 Resources 2 Timing and Newsletter Services Bob Brinker Timing Service Mutual Fund Portfolios somewhat conservative Timing Cube Daily timing service tuned for the QQQQ s but applicable to other indexes Country Fund ranking service NoLoad FundX Fund rankings in several volatility categories You could use this service for all the ranking Combined with your fund research using Appel s guidelines this could be a complete solution 03/06/

61 Financial Analyst Journal References 1. Chen, J., A. Bennet, T. Zheng, 2006, Sector Effects in Developed vs. Emerging Markets, Financial Analysts Journal, vol. 62, no. 6, (November/December): Scowcroft, A., J. Sefton, 2005, Understanding Momentum, Financial Analysts Journal, vol. 61, no. 2, (March/April): Desrosiers, S., J. L Her, J. Plante, 2004, Style Management in Equity Country Allocation, Financial Analysts Journal, vol. 60, no. 6, (November/December): Cavaglia, Stefano, and, V Moroz, 2002, Cross-Industry, Cross- Country Allocation, Financial Analysts Journal, vol.??, no.?, (November/December): O Neal, Edward, 2000, Industry Momentum and Sector Mutual Funds, Financial Analysts Journal, vol.??, no.?, (July/August): /06/

62 Industry ETF Rotation Models Instead of using Fidelity Sector funds we can use ETF s such as ishares used here. Due to data history limitations there are fewer international sectors here Performance Notes Performance no better than market style / cap segments I chose the funds with the longest possible data series. Newer funds have come along EFA and EEM international funds New global sector funds although some just started trading in October I believe this is a viable option but test funds with smaller amounts of historical data 03/06/

63 Industry ETF Rotation Models 03/06/

64 Computation and Statistical Significance of the alpha and beta estimates of Model 1

65 Statistical Significance Regression Model (Ind_Rot_Rtn 90d_Treas_Rtn) = alpha + beta x (VFINX_Rtn 90d_Treas_Rtn) + err Regression analysis Independent Variable excess weekly returns of VFINX over 90 day treasury bills Dependent Variable excess weekly returns of Industry Rotation over 90 day treasury bills Determine alpha and beta and their significance We would like to see a positive alpha and a beta less than 1.0 Positive alpha means our rotation algorithm is adding value Beta less than 1.0 means our returns are less dependent on market returns and may be less volatile 03/06/

66 Statistical Significance 2 (Ind_Rot_Rtn 90d_Treas_Rtn) = alpha + beta x (VFINX_Rtn 90d_Treas_Rtn) + err (Ind_Rot_Rtn 90d_Treas_Rtn) is plotted on Y Axis (VFINX_Rtn 90d_Treas_Rtn) is plotted on X Axis Alpha is the value of the Y axis intercept of pink line Beta is the slope of the pink line 03/06/

67 Statistical Significance 3 Alpha is found to be significant at 0.12% per week with a small standard error. Beta is found to be significant at 0.82 with a small standard error. R Square indicates that 75% of the weekly returns of our Industry Rotation model is explained by this model. Overall, these results are consistent with journal research results. 03/06/

68 Back-test Modeling Issues

69 Avoid Look-Ahead Errors A look-ahead error is a simulation trading on historical information that was not known at that date. If simulating January 1, 1995, you can only use data that you could have known and traded with on January 1, 1995 Many ways to incur this error in your simulation some obvious and some subtle Examples Put in a rule to avoid market crashes in 1997 and (obvious look-ahead error) Survivorship bias in your data (subtle look-ahead error) Failed companies have been removed from many data bases so your simulated historical trading cannot purchase these companies. Your algorithm looks much better than it really is! 03/06/

70 Avoid Curve Fitting Too many rules with too many tuned parameters can result in a model that is curve fit to the test data: Model performs brilliantly in the in-sample test data period but loses lots of money in the out-of-sample period with new data. Solutions use either or both Use a walk-forward development process Check the solution space for a robust flat plane area to locate your parameters Make sure you have enough data and simulated trades to support the number of rules and tuned parameters you use. Higher numbers of rules and tuned parameters require much more data and a much higher number of simulated trades Very old historical data may be less relevant to today s market Always a good idea: Keep models to a few simple rules. 03/06/