1 Harlyn Research Probability Based Investing Simon Goodfellow
1 Harlyn Research Probability. Based Investing Simon Goodfellow TELL ME WHY I DON’T LIKE FORECASTS (WITH APOLOGIES TO SIR BOB GELDOF AND THE BOOMTOWN RATS) +44 7710 123 588 SUPRA 2013 +44 176 320 8601 simon. goodfellow@ harlynresearch. com HILTON WALDORF LONDON www. harlynresearch. com 25 TH SEPTEMBER 2013 Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
Tell me why I don’t like…. 2 �Forecasts �Valuation �Mean Reversion �“Like” means we don’t think they are necessary for a successful investment process Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
What do we like? 3 �P �R �A �T �E �R Probability Risk. Adjusted Total Excess Return The PRATER process is at the heart of our approach to asset allocation. We apply it to multi-asset mandates, regional equity mandates and sector rotation models � Use historic total return data to calculate probability that one asset class will beat another on a risk-adjusted basis � All investment is based on probability but few organisations take such a direct approach Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
How to calculate the probability that equities will beat bonds over the next 12 months 4 US Equity & US Treasury returns: cumulative frequency distribution 100% 80% 60% � Use total return data has to be high-frequency mean and standard deviation no need for skew or kurtosis � Short data set (less than 52 weeks) Equity beats bonds long data sets include too much data which has become irrelevant � Multiple sample periods 40% Relative size of two areas shows prob. of equity beating 20% -4 0% -3 0% -2 0% -1 0% 0% 10 % 20 % 30 % 40 % 50 % 60 % 0% US Equity US Treasury Harlyn Capital: Private & confidential; restricted access can’t know which is the best predictor controls base effects � Take an average or base the portfolio on 20 different samples and allocate 5% to each 05/12/2020
Left shift: how to adjust the return distribution for the extra risk of equities 5 Cumulative frequency distribution with left shift � We want a risk-efficient portfolio � Set a hurdle rate � Equity risk premium = volatility of 100% equities minus volatility of bonds 80% specific to each sample period � Shift equity distribution to the left 60% Notice how prob. of equity beating bonds shrinks 40% 20% -4 0 -3 % 0% -2 0% -1 0% 0% 10 % 20 % 30 % 40 % 50 % 60 % 0% US Eq Bond by the amount of excess volatility � New, smaller area represents the risk-adjusted probability of equities beating bonds � Risk adjustment will vary according to market conditions US Eq - left shifted Harlyn Capital: Private & confidential; restricted access 05/12/2020
Risk conditions change. . . … quick responses are vital 6 � Rebalance once a week PRATER: Last five years US Equity vs US bonds � Equity weight = probability of 100% equities beating bonds 80% � Main drivers are XS return and XS volatility � No constraints 60% 40% 20% 3 /1 09 2 /1 09 1 /1 09 0 /1 09 9 /0 09 /0 8 0% 09 by definition Equity weight Harlyn Capital: Private & confidential; restricted access recommendations are already adjusted for risk can tailor output to include minimum or maximum position limits � Simple, transparent, responsive � See next page for results 05/12/2020
US Equity / Treasuries Probability Model Better risk-adjusted returns & smaller drawdown 7 Total return sinception: PRATER model vs Control 400% 477% sinception is equivalent to a CAGR of 9. 2% vs 7. 5% for Control Ave volatility sinception 18. 5% 20. 0% 15. 0% 10. 0% 6. 4% 9. 0% 5. 0% 0. 0% 200% US Equity Bond Control PRATER Max drawdown sinception 0. 0% 12 /9 12 5 /9 12 7 /9 12 9 /0 12 1 /0 12 3 /0 12 5 /0 12 7 /0 9 12 /1 1 100% US Equity Control Bond PRATER model can own any mix of US Equity & Bonds based on probability of risk-adjusted total excess return of equities vs bonds. Model rebalanced weekly. Control = constant 50/50. Harlyn Capital: Private & confidential; restricted access -9. 3% -20. 0% -25. 0% -40. 0% -60. 0% -11. 9% -54. 7% 05/12/2020
EM Equity / Treasuries Probability Model Better risk-adjusted returns & smaller drawdown 8 Total return sinception: PRATER model vs Control 800% 1050% sinception is equivalent to a CAGR of 14. 1%vs 7. 4% for Control Ave volatility sinception 30. 0% 23. 5% 20. 0% 12. 2% 11. 4% 400% 10. 0% 200% 0. 0% 6. 4% EM Equity 100% Bond Control PRATER Max drawdown sinception 0. 0% 12 /9 12 5 /9 12 7 /9 12 9 /0 12 1 /0 12 3 /0 12 5 /0 12 7 /0 9 12 /1 1 50% EM Equity Control Bond PRATER model can own any mix of EM Equity & Bonds based on probability of risk-adjusted total excess return of equities vs bonds. Model rebalanced weekly. Control = constant 50/50. Harlyn Capital: Private & confidential; restricted access -9. 3% -20. 0% -40. 0% -16. 1% -35. 9% -60. 0% -80. 0% -62. 8% 05/12/2020
US Hi Yield / Treasuries Probability Model Better risk-adjusted returns & smaller drawdown 9 Total return sinception: PRATER model vs Control Ave volatility sinception 10. 0% 8. 4% 400% 6. 4% 5. 7% 5. 4% 5. 0% 200% 0. 0% 100% 395% sinception is equivalent to a CAGR of 8. 0% vs 6. 1% for Control Max drawdown sinception 0. 0% 12 /9 12 5 /9 12 7 /9 12 9 /0 12 1 /0 12 3 /0 12 5 /0 12 7 /0 9 12 /1 1 50% US Hi Yield Bond Control PRATER US Hi Yield Control Bond PRATER model can own any mix of US Hi Yield & Bonds based on probability of risk-adjusted total excess return of equities vs bonds. Model rebalanced weekly. Control = constant 50/50. Harlyn Capital: Private & confidential; restricted access -9. 3% -20. 0% -40. 0% -14. 7% -9. 1% -33. 6% 05/12/2020
T AURA BE ST LIA LG RI BR IU A CA A M N ZIL CHADA CO C IL LOHI E MN DE CZ BIA N E A FI MACH N R L K F GE RAAN R N D H G MA CE ON R N E Y H GK EC UN O E N IN GA G DO IN RY D IR NE IA EL SI IS AN A RA D IT EL JA AL M KOPA Y AL R N N ET A E H ME YS A ER X IA I N LANCO OR D W S PA A KI NY PH S Z IL P TAN IP E P R P PO O INEU RTLA S UN RU GAD SI SA SS L N FR IA GA I PO CA SWSP RE EDAIN S E THTAIWISN AI WA S TULA N RKND EY U USK A AU S Basic idea tested in over 40 countries. . . 10 CAGR sinception: PRATER model vs control (all in US dollars vs US bonds) 20% Median increase in return is 3. 8% CONTROL Harlyn Capital: Private & confidential; restricted circulation. Median increase in volatility is 0. 4% 15% 10% 5% 0% MODEL 05/12/2020
Use conditional probability chain to create complex portfolios. Flexible modular system 11 Equity regions US Equity Select any number of equity regions EM Fixed income Japan Equity Euro Equity Other Eqty * Small Cap* Inv. Grade Corp US Treas. UK Equity Alternatives US High Yield Other FI * Commo dity EM Gov’t Bond Hedge Funds* Beware highly EM Other complex bond Equity portfolios Create portfolio REIT Alternatives only UK included in Equity extended model Other Equity HARLYN MULTIASSET * Assets marked with an asterisk are not included in any of the portfolios shown in this presentation Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
US Conventional Multi-Asset Model & Constituents Performance sinception (Jan ‘ 96) 12 Log scale US$ 742% 640% 320% 742% = 12. 0% CAGR vs 7. 2% for Control 333% 160% 80% Control = Constant 60% Global Equity 40% Bonds 12 /9 12 5 /9 6 12 /9 12 7 /9 8 12 /9 12 9 /0 0 12 /0 12 1 /0 12 2 /0 12 3 /0 12 4 /0 12 5 /0 6 12 /0 12 7 /0 12 8 /0 9 12 /1 0 12 /1 1 12 /1 2 40% Euro Eq UK Eq EM Eq US Eq EM Bond IG Bond Japan Eq US Treas Harlyn Control Harlyn Research LLP: Private & confidential; restricted circulation. Hi Yield 05/12/2020
US Extended Multi-Asset Model & Constituents Performance sinception (Jan ‘ 96) 13 Log scale US$ 1280% 640% 1256% = 15. 4% CAGR vs 7. 3% for Control. Note that Extended Model has occasional very high exposure to alternatives 354% 320% 160% 80% Euro Eq Hi Yld UK Eq IG Corp EM Eq US Treas Harlyn Research LLP: Private & confidential; restricted circulation. US Eq HARLYN 1 /1 12 12 /0 9 7 /0 12 5 /0 12 3 12 /0 /0 12 9 12 /9 7 /9 12 /9 5 12 Japan Eq EM Bond 1 Control = Constant 54% Global Equity, 6% Alternative, 40% Bonds 40% Commod Control REIT 05/12/2020
BACK TO THE RATS! 14 WHY DON’T WE LIKE WHAT CONVENTIONAL MANAGERS LOOK AT Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
Forecasts are not the answer 15 UK GDP forecast vs historic growth rate 6. 0 � Too many forecasts are 5. 0 wrong or already in the price 4. 0 � Many forecasts are not even 3. 0 forecasts (see chart) � Information overload 2. 0 1. 0 0. 0 -1. 0 -2. 0 * e. g. forecast for 2005 -3. 0 at the end of 2003 3 12 /0 5 12 /0 7 12 /0 9 12 /1 1 12 /1 3 /0 1 12 /0 9 12 12 /9 7 -4. 0 UK - Annual GDP growth rate the more forecasts you have, the longer it takes to change them. � More analysts ≠ better forecasts � More analysts = more incorrect forecasts � Forecasts get in the way of portfolio management Source: ING & Consensus Economics Harlyn Research LLP: Private & confidential; restricted circulation. RESPONSIVENESS IS ALL 05/12/2020
Nor are valuations 16 UK real gilt yields vs next 12 month real return 1. 00 0. 75 0. 50 0. 25 0. 00 -0. 25 -0. 50 � Valuation is an excellent tool for describing investments… �. . with little predictive power � Relationship between value and returns is fundamentally and permanently unstable � Chart shows UK real gilt yields (7 -10 year) vs real returns � Average correlation is +0. 22 -0. 75 0 /1 12 12 /0 6 8 � Markets do correct towards fair 4 2 12 /0 /0 12 /9 6 12 12 /9 8 Average = 0. 22 -1. 00 32% of observations are negative Correlation UK real yield and next 12 months real return Harlyn Research LLP: Private & confidential; restricted circulation. value EVENTUALLY WHEN RISK CONDITIONS ALLOW 05/12/2020
Risk offers a better explanation of returns 17 52 w correlation: XS equity vol. & XS equity returns 1. 00 Average = -0. 43 0. 75 � Chart shows correlation between 0. 50 � Average correlation is -0. 43 0. 25 0. 00 excess volatility (UK equity - gilts) excess return (UK equity - gilts) all based on nominal total returns much better than +0. 22 much tighter range � Persistent through time -0. 25 -0. 50 93% of observations are negative � In English this means -0. 75 12 /9 6 12 /9 8 12 /0 0 12 /0 2 12 /0 4 12 /0 6 12 /0 8 12 /1 0 12 /1 2 -1. 00 when equities get more risky (relative to bonds) their returns fall (relative to bonds) this happens most of the time Low XS risk: high XS return Harlyn Research LLP: Private & confidential; restricted circulation 05/12/2020
Mean reversion is an unreliable strategy 18 Total return sinception: Reverse PRATER model vs Control 400% 201% sinception is equivalent to a CAGR of 4. 0% vs 7. 5% for Control � Mean reversion is easy to observe in hindsight Much harder to predict in future Amplitude, frequency and trend are all unstable � Real problem is that it doesn’t work in portfolio management 200% THOUGHT EXPERIMENT � If mean reversion worked our 12 /9 12 5 /9 12 7 /9 12 9 /0 12 1 /0 12 3 /0 12 5 /0 12 7 /0 9 12 /1 1 100% US Equity Control Bond Reversion model can own any mix of US Equity & Bonds based on probability of risk-adjusted negative excess return of equities vs bonds. Control = constant 50/50. Harlyn Capital: Private & confidential; restricted access PRATER model should use high XS return to indicate future underperformance, not continuing outperformance � See slide 7 for comparison 05/12/2020
Probability vs Reversion Models vs 50/50 UK, Japan, Eurozone & Emerging Markets 19 UK: sinception 800% 400% Eurozone: sinception 400% 200% 100% 12/95 12/00 12/10 12/05 Japan: sinception 200% 12/95 50/50 12/05 Reversion 12/00 12/05 12/10 Emerging Markets: sinception 400% 50% 12/95 12/10 PRATER Harlyn Capital: Private & confidential; restricted access 800% 400% 200% 100% 50% 12/95 50/50 12/05 Reversion 12/10 PRATER 05/12/2020
Mean reversion style leads to high volatility 20 Annualised volatility sinception for selected mean reversion models 50/50 20. 0% Reversion 13% 10. 0% 9% USA 9% 12% 9% 8% UK PRATER 10% 12% 17% 15% 10% Japan 10% Eurozone 11% 12% Emerging � Our mean reversion models have exactly the same risk-adjustment process as our probability models � Returns are worse than 50/50 � Volatility is also much higher (drawdowns are also much worse) � Mean reversion models have to be constrained because they don’t work very well Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
Final Thoughts 21 � Forecasts: too many; too wrong; too often � Valuation: has less predictive power than volatility � Mean reversion: buying low is more risky than it sounds � Probability-based investing is simple, transparent & responsive automatically adjusts for risk produces higher returns and much smaller drawdowns � Modern risk management systems have evolved because the basic tools of investment analysis don’t work very well. Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
Disclaimer 22 This presentation has been prepared by Harlyn Research LLP (‘Harlyn’), and published by Fundamental Tracker Investment Management Limited (‘FTIM’), which is authorised and regulated by the FSA. This presentation does not constitute any offer or invitation to buy or subscribe or sell any share or security or other investment or to engage in any investment activity; neither does it constitute any advice to do so. Past performance is not a guarantee of future performance. The figures in this presentation are by way of illustration only. The illustrated returns are stated before transaction costs and management expenses and performance fees and are therefore not the returns that would have accrued to an investor if he had adopted the positions recommended by the model(s) during the period(s) indicated. It is not possible to estimate what such costs would be. There can be no guarantee that an investor would have been able to transact or to enter into transactions at the prices adopted by the model(s). No account has been taken in this presentation of the impact of any regulatory restriction or suspension of dealings that may occur in any market from time to time. Any investments referred to herein may involve significant risk, are not necessarily available in all jurisdictions, may be illiquid and may not be suitable for all investors. The value of, or the income from, any investments referred to herein may fluctuate and/or be affected by exchange rates. Investors should make their own investigations and investment decisions without relying on this presentation. Only investors with sufficient knowledge and experience in financial matters to evaluate the merits and risks should consider any investment in any issuer or market discussed herein and other persons should not take any action on the basis of this presentation. This presentation is issued in the United Kingdom only to persons described in Articles 19, 47 and 49 of the Financial Services Act 2000. Whilst reasonable care has been taken to ensure that the information contained in this presentation is not untrue or misleading at the time of publication, no representation or warranty, express or implied as to its completeness or its accuracy is made by Harlyn or FTIM or any of their officers employees and agents and no information contained in this presentation shall form the basis of any contract. The information in this presentation is subject to change without notice. Neither Harlyn nor FTIM nor any of their officers makes any commitment to update the information in this presentation or to correct any statement or figure in it or to correct, update or amend any underlying assumption or to provide any further information. Neither Harlyn, nor FTIM nor their respective officers or employees or agents accepts any liability for any direct or consequential loss arising from any use of this presentation. Copyright and database rights protection exists in this presentation and it may not be reproduced, distributed or published by any person for any purpose without the prior express written consent of Harlyn and FTIM. All rights are reserved. Harlyn Research LLP (registered in England Wales; number OC 369433); registered office: The Green High Street Fowlmere Royston SG 8 7 SS Fundamental Tracker Investment Management Limited (registered in Scotland, number SC 299736), registered office: Suite J, Exchange House, 50 Drymen Road, Bearsden, Glasgow G 61 2 RH Harlyn Research LLP: Private & confidential; restricted circulation. 05/12/2020
- Slides: 22