Bayesian Dynamic Linear Models for Strategic Asset Allocation
0 likes25 views
This document discusses Bayesian dynamic linear models for strategic asset allocation. It presents an approach using Bayesian modeling to predict excess returns on stocks and bonds based on predictor variables. The models allow for time-varying parameters and stochastic volatility. The approach averages predictions across multiple models to improve performance. It finds that accounting for parameter uncertainty and time-variation through Bayesian modeling and model averaging improves out-of-sample return and risk predictions compared to standard linear models without these features.
![Modeling Details
Prior created on 1962-1971 data
Models evaluated on 1985-2014
Evaluated on both economic and statistical criteria.
I Economic Measure: Certainty Equivalent Returns, using power utility
(CRRA)
F U(wealth) = 1
1−γ
(wealth)1−γ
F γ = 5
I Statistical Prediction Measure: Mean Squared Prediction Error Ratio
I Statistical Fit Measure: Average Log Score
Restrict:
I Portfolio weights wt ∈ [−2, 3]
I Coefficient variance discount factor δ ∈ [0.98, 1.0]
I Volatility discount factor δv ∈ [0.9, 1.0]
Fisher (UT) Bayesian Risk Prediction April 18, 2016 14 / 50](https://image.slidesharecdn.com/dlmassetallocation-240407124345-74b464c2/85/Bayesian-Dynamic-Linear-Models-for-Strategic-Asset-Allocation-14-320.jpg)
![Modeling Details
A model is fit for every combination of predictor, δ, and δv.
10 values of δ and δv are considered, equally spaced in the range
δ ∈ [0.98, 1.0], δv ∈ [0.9, 1.0], for a grid of 100 possibilities.
Fisher (UT) Bayesian Risk Prediction April 18, 2016 18 / 50](https://image.slidesharecdn.com/dlmassetallocation-240407124345-74b464c2/85/Bayesian-Dynamic-Linear-Models-for-Strategic-Asset-Allocation-18-320.jpg)
Bayesian Dynamic Linear Models for Strategic Asset Allocation
- 1. Bayesian Dynamic Linear Models for Strategic Asset Allocation Jared Fisher Carlos Carvalho, The University of Texas Davide Pettenuzzo, Brandeis University April 18, 2016 Fisher (UT) Bayesian Risk Prediction April 18, 2016 1 / 50
- 2. 1 Introduction 2 Single Risky Asset 3 Multiple Risky Assets 4 Conclusion Fisher (UT) Bayesian Risk Prediction April 18, 2016 2 / 50
- 3. Excess Returns on an Index: is there Signal in the Noise? 1960 1970 1980 1990 2000 2010 −20 −10 0 10 Year Percent Excess Return Stock Index 2 Year Bond Index 5 Year Bond Index Fisher (UT) Bayesian Risk Prediction April 18, 2016 3 / 50
- 4. How should an investor optimally create a portfolio? Two step process: Establish predictions of the mean and variance of assets’ future excess returns Use these estimates to determine how much of portfolio to devote to each asset. Return on a portfolio is a weighted sum of the individual assets’ returns, where the weights are the proportions invested. Fisher (UT) Bayesian Risk Prediction April 18, 2016 4 / 50
- 5. Making Investments Given forecasted µ̂t, Σ̂t For an investor with power utility and risk aversion γ Portfolio weights vector is wt = 1 γ Σ̂−1 t µ̂t + 1 2 diag(Σ̂t) Fisher (UT) Bayesian Risk Prediction April 18, 2016 5 / 50
- 6. Understanding Excess Returns What is the distribution of Yi,t+1 = (Ri,t+1 − Rf,t+1), given what we know at time t? E(Yi,t+1|Dt) (“risk premium”) I = µ? (constant, no predictability) I = µt = f(Xt) = X0 tβ? I = µt = X0 tβt? (“time-varying parameters”) V ar(Yi,t+1|Dt) I = σ2 ? (constant volatility) I = σ2 t ? (“stochastic volatility”) Fisher (UT) Bayesian Risk Prediction April 18, 2016 6 / 50
- 7. Does Predictability Exist? Literature assumes linear relationship: Yi,t+1 = X0 tβ + t+1, V ar() = σ2 Tests are mostly in-sample, not out-of-sample (OOS). Welch and Goyal (2008) show that the good performance of popular variables in-sample don’t hold OOS. More recently, authors show OOS predictability by deviating from the standard model. I Time-varying parameters (e.g. Dangl and Halling, 2012) I Stochastic volatility (e.g. Johannes, Korteweg and Polson, 2013) I Parameter uncertainty (Bayesian models) Fisher (UT) Bayesian Risk Prediction April 18, 2016 7 / 50
- 8. Our Analysis Two research questions I Predictability: is there useful information in X? I Time-variation: are the parameter values (β and σ2 ) constant with respect to time? Compare models with and without predictors and with and without variance discounting (of both regression coefficients and volatility) Benchmark: the constant model (i.e. Xt = 1) I Often called the expectation hypothesis model, it represents the efficient markets hypothesis/no predictability. Fisher (UT) Bayesian Risk Prediction April 18, 2016 8 / 50
- 9. Data description We will first look at portfolios of a risky asset (stock index or bond index) and a risk-free asset (3 month T-bill). We use the following data, spanning 1962-2014: Welch and Goyal’s predictors of stock performance, updated to 2014, CRSP value weighted returns, Bonds data from Gargano, Pettenuzzo, and Timmermann (2015) I Bond index for 2-5 year maturities, I Cochrane and Piazzesi’s (2005) linear combination of forward rates, I Fama and Bliss’ (1987) forward spread, I Ludvigson and Ng’s (2009) macro factor. Fisher (UT) Bayesian Risk Prediction April 18, 2016 9 / 50
- 10. Our Model Y 0 t = X0 t−1Bt + v0 t Bt = Bt−1 + Ωt vt ∼ N(0, VtΣt) Ωt ∼ N(0, Wt, Σt) (B0, Σ0|D0) ∼ NW−1 n0 (m0, C0, S0) Σt|Dt−1 ∼ W−1 δvnt−1 (St−1) Wt = 1 − δ δ Ct−1 Fisher (UT) Bayesian Risk Prediction April 18, 2016 10 / 50
- 11. Yt = X0 t−1βt + t 1980 1990 2000 2010 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 Year β DY OLS-Expanding Window Dyamic Linear Model DLM with Time-varying Parameters Full-term OLS Fisher (UT) Bayesian Risk Prediction April 18, 2016 11 / 50
- 12. Model Advantages Bayesian model without need of MCMC I Allows us to fits more models in the same amount of computation time Bridges gap between Recursive model vs. Rolling-window model Fisher (UT) Bayesian Risk Prediction April 18, 2016 12 / 50
- 13. Recursion . . . βt−1, Σt−1|Dt−1 ∼ NW−1 nt−1 (mt−1, Ct−1, St−1) βt, Σt−1|Dt−1 ∼ NW−1 nt−1 (mt−1, Rt, St−1) Rt = Ct−1 + Wt = 1 δ Ct−1 Yt|Dt−1 ∼ Tnt−1 (X0 t−1mt−1, QtSt−1) Qt = Vt + X0 t−1RtXt−1 βt, Σt|Dt ∼ NW−1 nt (mt, Ct, St) et = Yt − X0 t−1mt−1 . . . At = RtXt−1/Qt nt = δvnt−1 + 1 mt = mt−1 + Ate0 t Ct = Rt − AtA0 tQt St = n−1 t δvnt−1St−1 + ete0 t Qt Fisher (UT) Bayesian Risk Prediction April 18, 2016 13 / 50
- 14. Modeling Details Prior created on 1962-1971 data Models evaluated on 1985-2014 Evaluated on both economic and statistical criteria. I Economic Measure: Certainty Equivalent Returns, using power utility (CRRA) F U(wealth) = 1 1−γ (wealth)1−γ F γ = 5 I Statistical Prediction Measure: Mean Squared Prediction Error Ratio I Statistical Fit Measure: Average Log Score Restrict: I Portfolio weights wt ∈ [−2, 3] I Coefficient variance discount factor δ ∈ [0.98, 1.0] I Volatility discount factor δv ∈ [0.9, 1.0] Fisher (UT) Bayesian Risk Prediction April 18, 2016 14 / 50
- 15. Results: No Discounting - Comparison to Literature Stock Index Bond Index Predictor CER ALS MSE Mat. Pred. CER ALS MSE (none) 5.678 1.689 1.000 2 (none) 6.519 3.512 1.000 Log D/P 3.385 1.681 1.018 2 CP 5.838 3.517 1.049 Log D/Y 3.333 1.682 1.018 2 FB 6.902 3.519 0.979 Log E/P 3.884 1.685 1.012 2 LN 8.535 3.531 1.010 Smooth E/P 3.228 1.680 1.019 3 (none) 6.265 3.146 1.000 Log D/Payout 0.767 1.681 1.020 3 CP 5.853 3.150 1.028 B/M 3.133 1.680 1.020 3 FB 7.614 3.154 0.975 T Bill Rate 5.424 1.687 1.003 3 LN 9.463 3.165 0.981 LngTerm Yld 5.507 1.688 1.002 4 (none) 6.083 2.891 1.000 LngTerm Ret. 4.630 1.686 1.007 4 CP 5.827 2.895 1.015 Term Spread 2.764 1.683 1.010 4 FB 8.192 2.901 0.974 Def.Yld.Sprd 1.854 1.678 1.023 4 LN 9.558 2.910 0.970 Def.Ret.Sprd 4.199 1.692 0.999 5 (none) 5.910 2.694 1.000 Stock Var. 6.426 1.701 0.977 5 CP 5.882 2.697 1.007 Net Eqty Exp. 4.288 1.682 1.013 5 FB 8.416 2.704 0.974 Inflation 2.980 1.683 1.012 5 LN 9.257 2.713 0.965 Fisher (UT) Bayesian Risk Prediction April 18, 2016 15 / 50
- 16. Discount Factor Heatmap - Grid of δ, δv 4.76 4.78 4.80 4.82 0.980 0.985 0.990 0.995 1.000 0.90 0.92 0.94 0.96 0.98 1.00 δ δ v LN 2 years CER Fisher (UT) Bayesian Risk Prediction April 18, 2016 16 / 50
- 17. Average Over Models Many models beat the benchmark, given the correct discount factors. But, we don’t know a priori how much to discount or which predictors will perform well. Solution: average and share strength across models. I For each time t, weight each of the models’ prediction based on its performance up through time t − 1. I Create different averaged models by weighting on utility and score, as well as an equal-weighted model. wU i,τ+1 = 1 γ 1 τ τ X t=1 Ui,t ! 1 1−γ wS i,τ+1 = τ X t=1 ln(scorei,t) ! − min j τ X t=1 ln(scorej,t) ! Fisher (UT) Bayesian Risk Prediction April 18, 2016 17 / 50
- 18. Modeling Details A model is fit for every combination of predictor, δ, and δv. 10 values of δ and δv are considered, equally spaced in the range δ ∈ [0.98, 1.0], δv ∈ [0.9, 1.0], for a grid of 100 possibilities. Fisher (UT) Bayesian Risk Prediction April 18, 2016 18 / 50
- 19. Model Averaging Results: Stocks Pred TVP SV Models Weights CER ALS MSE 0 0 0 1 (none) 5.678 1.689 1.0000 0 0 1 10 Equal 5.904 1.718 1.0000 0 0 1 10 Utility 5.904 1.718 1.0000 0 0 1 10 Score 6.182 1.725 1.0000 0 1 0 10 Equal 5.458 1.689 1.0005 0 1 0 10 Utility 5.458 1.689 1.0005 0 1 0 10 Score 4.914 1.687 1.0034 0 1 1 100 Equal 5.717 1.717 1.0005 0 1 1 100 Utility 5.717 1.717 1.0005 0 1 1 100 Score 5.750 1.723 1.0009 1 0 0 16 Equal 5.787 1.693 0.9992 1 0 0 16 Utility 5.786 1.693 0.9992 1 0 0 16 Score 6.060 1.695 0.9967 1 0 1 160 Equal 5.906 1.721 0.9992 1 0 1 160 Utility 5.906 1.721 0.9992 1 0 1 160 Score 6.306 1.725 0.9968 1 1 0 160 Equal 4.984 1.697 0.9999 1 1 0 160 Utility 4.984 1.697 0.9999 1 1 0 160 Score 4.857 1.697 0.9991 1 1 1 1600 Equal 5.400 1.722 0.9999 1 1 1 1600 Utility 5.400 1.722 0.9999 1 1 1 1600 Score 5.670 1.724 0.9982 Fisher (UT) Bayesian Risk Prediction April 18, 2016 19 / 50
- 20. Model Averaging Results: Stocks ● ● ● ● 1.69 1.70 1.71 1.72 1.73 1.74 5.0 5.5 6.0 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted Stocks Fisher (UT) Bayesian Risk Prediction April 18, 2016 20 / 50
- 21. Model Averaging Results: Bonds, 2 Year Maturity Pred TVP SV Models Weights CER ALS MSE 0 0 0 1 (none) 6.519 3.512 1.0000 0 0 1 10 Equal 7.637 3.784 1.0000 0 0 1 10 Utility 7.637 3.784 1.0000 0 0 1 10 Score 7.716 3.835 1.0000 0 1 0 10 Equal 7.355 3.512 0.9907 0 1 0 10 Utility 7.355 3.512 0.9907 0 1 0 10 Score 7.210 3.512 0.9957 0 1 1 100 Equal 7.953 3.784 0.9907 0 1 1 100 Utility 7.953 3.784 0.9907 0 1 1 100 Score 8.088 3.834 0.9908 1 0 0 4 Equal 7.651 3.522 0.9652 1 0 0 4 Utility 7.653 3.522 0.9652 1 0 0 4 Score 8.013 3.528 0.9829 1 0 1 40 Equal 8.261 3.788 0.9652 1 0 1 40 Utility 8.262 3.788 0.9652 1 0 1 40 Score 8.308 3.837 0.9641 1 1 0 40 Equal 7.987 3.522 0.9650 1 1 0 40 Utility 7.988 3.522 0.9650 1 1 0 40 Score 8.199 3.528 0.9631 1 1 1 400 Equal 8.254 3.792 0.9650 1 1 1 400 Utility 8.255 3.792 0.9650 1 1 1 400 Score 8.247 3.840 0.9644 Fisher (UT) Bayesian Risk Prediction April 18, 2016 21 / 50
- 22. Model Averaging Results: Bonds, 2 Year Maturity ● ● ● ● 3.5 3.6 3.7 3.8 3.9 4.0 6.5 7.0 7.5 8.0 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted Bonds_2 Fisher (UT) Bayesian Risk Prediction April 18, 2016 22 / 50
- 23. Model Averaging Results: Bonds, 3 Year Maturity Pred TVP SV Models Weights CER ALS MSE 0 0 0 1 (none) 6.265 3.146 1.0000 0 0 1 10 Equal 7.685 3.320 1.0000 0 0 1 10 Utility 7.685 3.320 1.0000 0 0 1 10 Score 7.908 3.343 1.0000 0 1 0 10 Equal 7.189 3.146 0.9922 0 1 0 10 Utility 7.189 3.146 0.9922 0 1 0 10 Score 7.019 3.146 0.9967 0 1 1 100 Equal 8.632 3.321 0.9922 0 1 1 100 Utility 8.632 3.321 0.9922 0 1 1 100 Score 8.693 3.343 0.9923 1 0 0 4 Equal 7.922 3.156 0.9686 1 0 0 4 Utility 7.925 3.156 0.9685 1 0 0 4 Score 8.677 3.161 0.9734 1 0 1 40 Equal 9.191 3.331 0.9686 1 0 1 40 Utility 9.193 3.331 0.9685 1 0 1 40 Score 9.365 3.354 0.9680 1 1 0 40 Equal 8.425 3.155 0.9699 1 1 0 40 Utility 8.427 3.155 0.9698 1 1 0 40 Score 8.913 3.159 0.9672 1 1 1 400 Equal 9.395 3.330 0.9699 1 1 1 400 Utility 9.396 3.330 0.9698 1 1 1 400 Score 9.437 3.352 0.9697 Fisher (UT) Bayesian Risk Prediction April 18, 2016 23 / 50
- 24. Model Averaging Results: Bonds, 3 Year Maturity ● ● ● ● 3.15 3.20 3.25 3.30 3.35 3.40 3.45 6.5 7.0 7.5 8.0 8.5 9.0 9.5 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted Bonds_3 Fisher (UT) Bayesian Risk Prediction April 18, 2016 24 / 50
- 25. Model Averaging Results: Bonds, 4 Year Maturity Pred TVP SV Models Weights CER ALS MSE 0 0 0 1 (none) 6.083 2.891 1.0000 0 0 1 10 Equal 7.461 3.009 1.0000 0 0 1 10 Utility 7.462 3.009 1.0000 0 0 1 10 Score 7.845 3.021 1.0000 0 1 0 10 Equal 7.002 2.892 0.9929 0 1 0 10 Utility 7.002 2.892 0.9929 0 1 0 10 Score 6.822 2.892 0.9972 0 1 1 100 Equal 8.226 3.010 0.9929 0 1 1 100 Utility 8.226 3.010 0.9929 0 1 1 100 Score 8.596 3.021 0.9930 1 0 0 4 Equal 7.972 2.902 0.9696 1 0 0 4 Utility 7.975 2.902 0.9695 1 0 0 4 Score 8.876 2.908 0.9681 1 0 1 40 Equal 9.433 3.023 0.9696 1 0 1 40 Utility 9.437 3.023 0.9695 1 0 1 40 Score 9.709 3.035 0.9693 1 1 0 40 Equal 8.293 2.901 0.9721 1 1 0 40 Utility 8.295 2.901 0.9720 1 1 0 40 Score 8.766 2.904 0.9689 1 1 1 400 Equal 9.583 3.020 0.9721 1 1 1 400 Utility 9.585 3.020 0.9720 1 1 1 400 Score 9.756 3.032 0.9721 Fisher (UT) Bayesian Risk Prediction April 18, 2016 25 / 50
- 26. Model Averaging Results: Bonds, 4 Year Maturity ● ● ● ● 2.90 2.95 3.00 3.05 3.10 6 7 8 9 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted Bonds_4 Fisher (UT) Bayesian Risk Prediction April 18, 2016 26 / 50
- 27. Model Averaging Results: Bonds, 5 Year Maturity Pred TVP SV Models Weights CER ALS MSE 0 0 0 1 (none) 5.910 2.694 1.0000 0 0 1 10 Equal 6.971 2.780 1.0000 0 0 1 10 Utility 6.971 2.780 1.0000 0 0 1 10 Score 7.330 2.786 1.0000 0 1 0 10 Equal 6.803 2.695 0.9934 0 1 0 10 Utility 6.803 2.695 0.9934 0 1 0 10 Score 6.597 2.694 0.9977 0 1 1 100 Equal 7.809 2.781 0.9934 0 1 1 100 Utility 7.809 2.781 0.9934 0 1 1 100 Score 8.025 2.786 0.9934 1 0 0 4 Equal 7.920 2.706 0.9697 1 0 0 4 Utility 7.921 2.706 0.9696 1 0 0 4 Score 8.620 2.710 0.9642 1 0 1 40 Equal 9.031 2.796 0.9697 1 0 1 40 Utility 9.034 2.796 0.9697 1 0 1 40 Score 9.470 2.803 0.9697 1 1 0 40 Equal 8.055 2.704 0.9734 1 1 0 40 Utility 8.057 2.704 0.9734 1 1 0 40 Score 8.380 2.706 0.9705 1 1 1 400 Equal 9.259 2.793 0.9734 1 1 1 400 Utility 9.262 2.793 0.9734 1 1 1 400 Score 9.513 2.799 0.9737 Fisher (UT) Bayesian Risk Prediction April 18, 2016 27 / 50
- 28. Model Averaging Results: Bonds, 5 Year Maturity ● ● ● ● 2.70 2.75 2.80 2.85 6 7 8 9 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted Bonds_5 Fisher (UT) Bayesian Risk Prediction April 18, 2016 28 / 50
- 29. Conclusions on Single Asset Models The best single risky asset models include predictors and stochastic volatility, perhaps with time-varying parameters for bonds. Does predictability exist? Yes, the best averaged model in most cases include predictors. Is time variation important? Yes, especially stochastic volatility. Fisher (UT) Bayesian Risk Prediction April 18, 2016 29 / 50
- 30. Our Multivariate Model Ideal portfolio probably contains more than one risky asset. Use this same model, but fit for multiple risky assets. Portfolio of the stock index and a bond index, for a given maturity. Each model can include one stock predictor and one bond predictor Fisher (UT) Bayesian Risk Prediction April 18, 2016 30 / 50
- 31. Multivariate Model Averaging Results, 2 year maturity Pred TVP SV Models Weights CER ALS MSE S. MSE B. 0 0 0 1 (none) 7.970 5.189 1.0000 1.0000 0 0 1 10 Equal 8.276 5.529 1.0000 1.0000 0 0 1 10 Utility 8.275 5.529 1.0000 1.0000 0 0 1 10 Score 6.469 5.578 1.0000 1.0000 0 1 0 10 Equal 8.420 5.189 1.0005 0.9907 0 1 0 10 Utility 8.420 5.189 1.0005 0.9907 0 1 0 10 Score 8.328 5.189 1.0011 0.9939 0 1 1 100 Equal 8.862 5.527 1.0005 0.9907 0 1 1 100 Utility 8.861 5.527 1.0005 0.9907 0 1 1 100 Score 7.135 5.575 1.0005 0.9908 1 0 0 64 Equal 9.105 5.202 1.0014 0.9607 1 0 0 64 Utility 9.107 5.202 1.0014 0.9606 1 0 0 64 Score 9.682 5.205 1.0000 0.9538 1 0 1 640 Equal 7.916 5.523 1.0014 0.9607 1 0 1 640 Utility 7.919 5.523 1.0014 0.9606 1 0 1 640 Score 6.244 5.572 1.0008 0.9589 1 1 0 640 Equal 7.636 5.208 1.0017 0.9593 1 1 0 640 Utility 7.637 5.208 1.0017 0.9593 1 1 0 640 Score 8.007 5.209 1.0022 0.9530 1 1 1 6400 Equal 8.355 5.533 1.0017 0.9593 1 1 1 6400 Utility 8.357 5.533 1.0017 0.9593 1 1 1 6400 Score 7.296 5.577 1.0012 0.9582 Fisher (UT) Bayesian Risk Prediction April 18, 2016 31 / 50
- 32. Multivariate Model Averaging Results, 2 year maturity ● ● ● ● 5.2 5.3 5.4 5.5 5.6 5.7 6.5 7.0 7.5 8.0 8.5 9.0 9.5 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted StocksBonds_2 Fisher (UT) Bayesian Risk Prediction April 18, 2016 32 / 50
- 33. Multivariate Model Averaging Results, 3 year maturity Pred TVP SV Models Weights CER ALS MSE S. MSE B. 0 0 0 1 (none) 7.749 4.822 1.0000 1.0000 0 0 1 10 Equal 9.137 5.066 1.0000 1.0000 0 0 1 10 Utility 9.137 5.066 1.0000 1.0000 0 0 1 10 Score 7.971 5.088 1.0000 1.0000 0 1 0 10 Equal 8.266 4.822 1.0005 0.9922 0 1 0 10 Utility 8.266 4.822 1.0005 0.9922 0 1 0 10 Score 8.047 4.822 1.0016 0.9955 0 1 1 100 Equal 9.799 5.064 1.0005 0.9922 0 1 1 100 Utility 9.798 5.064 1.0005 0.9922 0 1 1 100 Score 8.734 5.085 1.0005 0.9923 1 0 0 64 Equal 9.590 4.836 1.0012 0.9644 1 0 0 64 Utility 9.593 4.836 1.0012 0.9644 1 0 0 64 Score 10.078 4.839 0.9997 0.9593 1 0 1 640 Equal 9.300 5.071 1.0012 0.9644 1 0 1 640 Utility 9.304 5.071 1.0012 0.9644 1 0 1 640 Score 7.849 5.096 1.0005 0.9635 1 1 0 640 Equal 7.875 4.840 1.0016 0.9656 1 1 0 640 Utility 7.877 4.841 1.0016 0.9655 1 1 0 640 Score 8.213 4.841 1.0021 0.9608 1 1 1 6400 Equal 9.321 5.073 1.0016 0.9656 1 1 1 6400 Utility 9.324 5.073 1.0016 0.9655 1 1 1 6400 Score 8.475 5.093 1.0010 0.9648 Fisher (UT) Bayesian Risk Prediction April 18, 2016 33 / 50
- 34. Multivariate Model Averaging Results, 3 year maturity ● ● ● ● 4.9 5.0 5.1 5.2 8.0 8.5 9.0 9.5 10.0 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted StocksBonds_3 Fisher (UT) Bayesian Risk Prediction April 18, 2016 34 / 50
- 35. Multivariate Model Averaging Results, 4 year maturity Pred TVP SV Models Weights CER ALS MSE S. MSE B. 0 0 0 1 (none) 7.561 4.568 1.0000 1.0000 0 0 1 10 Equal 9.610 4.755 1.0000 1.0000 0 0 1 10 Utility 9.609 4.755 1.0000 1.0000 0 0 1 10 Score 8.916 4.764 1.0000 1.0000 0 1 0 10 Equal 8.060 4.568 1.0005 0.9929 0 1 0 10 Utility 8.060 4.568 1.0005 0.9929 0 1 0 10 Score 7.734 4.567 1.0020 0.9974 0 1 1 100 Equal 10.002 4.753 1.0005 0.9929 0 1 1 100 Utility 10.002 4.753 1.0005 0.9929 0 1 1 100 Score 9.441 4.762 1.0005 0.9930 1 0 0 64 Equal 9.704 4.583 1.0012 0.9660 1 0 0 64 Utility 9.707 4.583 1.0012 0.9660 1 0 0 64 Score 10.017 4.585 0.9998 0.9623 1 0 1 640 Equal 9.705 4.766 1.0012 0.9660 1 0 1 640 Utility 9.710 4.766 1.0012 0.9660 1 0 1 640 Score 8.677 4.779 1.0004 0.9655 1 1 0 640 Equal 7.661 4.587 1.0015 0.9690 1 1 0 640 Utility 7.663 4.587 1.0015 0.9689 1 1 0 640 Score 7.968 4.587 1.0020 0.9650 1 1 1 6400 Equal 9.722 4.764 1.0015 0.9690 1 1 1 6400 Utility 9.728 4.764 1.0015 0.9689 1 1 1 6400 Score 9.238 4.774 1.0009 0.9684 Fisher (UT) Bayesian Risk Prediction April 18, 2016 35 / 50
- 36. Multivariate Model Averaging Results, 4 year maturity ● ● ● ● 4.60 4.65 4.70 4.75 4.80 4.85 7.5 8.0 8.5 9.0 9.5 10.0 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted StocksBonds_4 Fisher (UT) Bayesian Risk Prediction April 18, 2016 36 / 50
- 37. Multivariate Model Averaging Results, 5 year maturity Pred TVP SV Models Weights CER ALS MSE S. MSE B. 0 0 0 1 (none) 7.366 4.371 1.0000 1.0000 0 0 1 10 Equal 9.294 4.525 1.0000 1.0000 0 0 1 10 Utility 9.295 4.525 1.0000 1.0000 0 0 1 10 Score 9.355 4.528 1.0000 1.0000 0 1 0 10 Equal 7.826 4.372 1.0005 0.9934 0 1 0 10 Utility 7.825 4.372 1.0005 0.9934 0 1 0 10 Score 7.449 4.370 1.0021 0.9985 0 1 1 100 Equal 9.831 4.524 1.0005 0.9934 0 1 1 100 Utility 9.831 4.524 1.0005 0.9934 0 1 1 100 Score 9.489 4.526 1.0005 0.9934 1 0 0 64 Equal 9.582 4.387 1.0012 0.9669 1 0 0 64 Utility 9.585 4.387 1.0012 0.9668 1 0 0 64 Score 9.793 4.389 1.0000 0.9643 1 0 1 640 Equal 9.782 4.540 1.0012 0.9669 1 0 1 640 Utility 9.786 4.540 1.0012 0.9668 1 0 1 640 Score 8.842 4.547 1.0004 0.9667 1 1 0 640 Equal 7.232 4.390 1.0016 0.9719 1 1 0 640 Utility 7.235 4.390 1.0016 0.9719 1 1 0 640 Score 7.462 4.390 1.0020 0.9689 1 1 1 6400 Equal 9.236 4.537 1.0016 0.9719 1 1 1 6400 Utility 9.242 4.537 1.0016 0.9719 1 1 1 6400 Score 9.027 4.542 1.0009 0.9716 Fisher (UT) Bayesian Risk Prediction April 18, 2016 37 / 50
- 38. Multivariate Model Averaging Results, 5 year maturity ● ● ● ● 4.40 4.45 4.50 4.55 4.60 7.5 8.0 8.5 9.0 9.5 ALS CER ● ● Benchmark SV TVP TVP−SV w/Predictors Equal−Weighted Utility−Weighted Score−Weighted StocksBonds_5 Fisher (UT) Bayesian Risk Prediction April 18, 2016 38 / 50
- 39. Summary The best single risky asset models include predictors and stochastic volatility, perhaps with time-varying parameters for bonds. If optimizing statistical fit (ALS), the best models of multiple risky assets include stochastic volatility, usually with predictors. If optimizing economic significance (CER), the best models of multiple risky assets include I Predictors alone for shorter maturities. I Time-varying parameters and stochastic volatility with no predictors for larger maturities, equal or utility weighted (also the balanced choice). Fisher (UT) Bayesian Risk Prediction April 18, 2016 39 / 50
- 40. Limitations The literature has shown that the time period used affects results. However, showing that there is predictability from 1985-2014 runs against Welch and Goyal’s finding that predictability disappears in the more recent data. Fisher (UT) Bayesian Risk Prediction April 18, 2016 40 / 50
- 41. Conclusions We demonstrate a Bayesian methodology that can quickly estimate a time-series model without requiring MCMC or another computation-intensive sampling algorithm. Time-varying parameters, stochastic volatility, and predictors generally show improvements over the benchmark model. Does predictability exist? Yes, the best averaged model in most cases include predictors. Is time variation important? Yes, especially stochastic volatility. Fisher (UT) Bayesian Risk Prediction April 18, 2016 41 / 50
- 42. Questions, Comments? Thank you! Fisher (UT) Bayesian Risk Prediction April 18, 2016 42 / 50
- 43. Different Risk Aversion What if γ = 10? Fisher (UT) Bayesian Risk Prediction April 18, 2016 43 / 50
- 44. Multivariate Portfolio Weights, 2 Year Maturity, γ = 5 1980 1990 2000 2010 -2 -1 0 1 2 3 Year Weight - Percent Invested Historic Mean Model - Portfolio Weights Stocks Bonds Start Eval Fisher (UT) Bayesian Risk Prediction April 18, 2016 44 / 50
- 45. Multivariate Portfolio Weights, 2 Year Maturity, γ = 10 1980 1990 2000 2010 -2 -1 0 1 2 3 Year Weight - Percent Invested Historic Mean Model - Portfolio Weights Stocks Bonds Start Eval Fisher (UT) Bayesian Risk Prediction April 18, 2016 45 / 50
- 46. Multivariate Portfolio Weights, 2 Year Maturity, γ = 5 1980 1990 2000 2010 -2 -1 0 1 2 3 Year Weight - Percent Invested Score-weighted Model, no Discounting - Portfolio Weights Stocks Bonds Start Eval Fisher (UT) Bayesian Risk Prediction April 18, 2016 46 / 50
- 47. Multivariate Portfolio Weights, 2 Year Maturity, γ = 10 1980 1990 2000 2010 -2 -1 0 1 2 3 Year Weight - Percent Invested Score-weighted Model, no Discounting - Portfolio Weights Stocks Bonds Start Eval Fisher (UT) Bayesian Risk Prediction April 18, 2016 47 / 50
- 48. Multivariate Portfolio Weights, 2 Year Maturity, γ = 5 1980 1990 2000 2010 -2 -1 0 1 2 3 Year Weight - Percent Invested Score-weighted Model, with Discounting - Portfolio Weights Stocks Bonds Start Eval Fisher (UT) Bayesian Risk Prediction April 18, 2016 48 / 50
- 49. Multivariate Portfolio Weights, 2 Year Maturity, γ = 10 1980 1990 2000 2010 -2 -1 0 1 2 3 Year Weight - Percent Invested Score-weighted Model, with Discounting - Portfolio Weights Stocks Bonds Start Eval Fisher (UT) Bayesian Risk Prediction April 18, 2016 49 / 50
- 50. Intervention Intervention: Expected risk premium should be non-negative if not positive. Fisher (UT) Bayesian Risk Prediction April 18, 2016 50 / 50