Not logged in.
Quick Search - Contribution
|Title||Forecasting Long-Term Expected Returns Based on Earnings - An Analysis of a Financial Industry Approach|
|Institution||University of Zurich|
|Faculty||Faculty of Business, Economics and Informatics|
|Number of Pages||72|
|Zusammenfassung||Financial markets have been exposed to recurring turmoil in the new millennium which triggered plummeted sovereign bond yields. These circumstances force institutional investors such as pension funds, to revise their strategic asset allocations which may determine approximately 90% of a portfolio's investment performance (Brinson, Singer, and Beebower (1991)). A pension fund's strategic asset allocation depends on a broad variety of factors, however, the most essential input parameters probably are expected returns and risk forecasts. This thesis is motivated by a financial industry approach developed by a Swiss consultancy which among other factors estimates five-year expected returns based on earnings across four different asset classes: stocks, bonds, real estate, and alternative investments. These return forecasts are decisive for asset liability management and strategic asset allocations of many Swiss pension funds. Thus, a detailed analysis of the earnings-based forecasting methodology is of great interest, as it differs from many other forecasting approaches which for example are based on dividends, forward looking earnings, or time series analysis. This thesis lays its focus on the empirical backtesting of this financial industry approach, as there seems to be sparse recent research regarding the formation of expected returns based on earnings across various asset classes. The work by Koijen, Moskowitz, Pedersen, and Vrugt (2015) seems to be the most regarded paper in the recent past, which discusses the ideas closest to the mentioned financial industry approach. They present empirical evidence that an asset's expected return assuming that market conditions stay constant, or as they call it "carry", predicts returns for different major asset classes. Hence, the objective of the thesis is to analyze the forecasting ability and accuracy of the forecasting approach mentioned above, where expected stock returns are generated by earnings yields and expected bond returns are estimated with the yield-to-worst. In a first step, the theory behind the forecasting approach and its exact calculation methodology is presented for each asset class, as there are various possibilities how these two variables can be calculated. In a second step, the accuracy of expected return forecasts of stocks and bonds is analyzed separately. This empirical analysis is undertaken on the basis of a long-term data set and a short-term data set for both asset classes. Finally, the expected return indicators are used to devise hypothetical investment strategies. The examination of the Standard & Poor's 500 index from 1871 to 2016 indicates that the stock forecast model based on earnings is a significant predictor of stock returns for a five-year horizon, but the expected returns underestimate the effective realized returns on average. Nonetheless, the model performs superior over a ten-year time interval in comparison to a five-year forecasting horizon. Expected stock returns have also been estimated for five different MSCI equity indices for the period from 1996 to 2016. The results show that the model's performance can significantly vary between the different markets. While the results of the North America index are comparable to findings of the long-term data set, the model overestimates the subsequent return over a five-year horizon for the World, Europe, Pacific, and Emerging Markets index. The analysis of the long-term U.S. government bond index from 1926 to 2015 implies that the yield-to-worst is a reasonable predictor of subsequent bond returns. If the forecasting horizon is close to the index's duration, the forecast error is relatively small, but the yield-to-worst can also considerably differ from the realized return in a worst case scenario. Although the yield-to-worst is a significant predictor of expected bond returns, it is demanding to determine how much of the subsequent return's variability is explained by the yield-to-worst, due to statistical issues. Furthermore, the exact financial industry approach has been examined for the Barclays U.S. Treasury index for the period from 1988 to 2016. This analysis reveals that the duration-weighted portfolio yield applied by the consultancy whose approach is analyzed within this thesis generates smaller forecast errors in comparison with the market-capitalization-weighted portfolio yield. When implementing the forecast measures in a portfolio context, the results are ambiguous: While the relative carry indicator appears to be unsuitable to forecast subsequent returns across stocks and bonds, an absolute carry indicator is able to generate a considerable annualized excess return compared to a 60/40 benchmark portfolio. The absolute carry strategy can identify the asset class with the higher risk premium, but the strategies' standard deviation increase as well, thus, the Sharpe ratio slightly decreases. The introduction of a portfolio carry which measures the expected return of a risky 60/40 portfolio compared with a risk-free asset generates more promising results. The strategy is invested in a risky portfolio if the portfolio carry is above the three-month T-Bill rate and is fully invested in cash if the risk-free asset promises more return. This strategy outperforms the benchmark portfolio and generates 0.7% annualized excess return after transaction costs during a period of 90 years. Compared to the absolute carry predictor, the return per unit of risk measured with the standard deviation increases likewise. Furthermore, the strategy never signals a trade which has a considerable negative impact on the portfolio value, but might not outperform the benchmark for many years. Nevertheless, the strategy is capable of delivering a positive return contribution to the portfolio's performance. Additionally, the strategy appears quite robust to parameter changes and guidelines of Bailey, Borwein, de Prado, and Zhu (2014) have been considered to reduce the probability of overfitting while developing these strategies. Overall, the results presented in this thesis indicate that long-term expected returns for stocks and bonds significantly predict subsequent returns for a five-year horizon, thus, the earnings-based measures can give beneficial investment advice for devising a strategic asset allocation. Nevertheless, an investor needs to be aware that the volatility of the forecast errors can be enormous and can cause a weak performance over many years. The results of this thesis are exposed to the power of the statistical models applied in this thesis. If the imposed assumptions, such as the inexistence of unit roots or none-endogeneity problems are violated, the statistical inference might be spurious and can lead to erroneous conclusions. To resolve some of these statistical issues further academic research is required. Nevertheless, the backtesting of the financial industry approach analyzed within this thesis will become more meaningful, as longer data samples, which mitigate the small sample issue of non-overlapping five-year observations, will be available in the future. Additionally, the robustness of the results could also be verified by examining the model's predictive power in out-of-sample regressions, but one needs to be aware that the historical relationships discussed throughout this thesis are not necessarily applicable to the future. This thesis gives valuable information to the financial industry approach applied in practice, but since the statistical methods used in this thesis are relatively straightforward and based on strong underlying assumptions, there is potential for the implementation of more advanced statistical methods. Furthermore, the financial consultancy whose approach was analyzed also forecasts expected returns for the asset classes of real estate and alternative investments based on earnings, which could be considered in future research. Generally, the topic of the predictability of asset returns is a research field of great interest, as it is open for new findings and motivates for further academic research.|