DIVERSIFICACIÓN DE INGRESOS EN LA BANCA
1.3. LOS INTERMEDIARIOS FINANCIEROS Y LA DIVERSIFICACIÓN DE INGRESOS: ENTORNO ACTUAL
In addition to the performance of various models, a few minor conclusions are drawn from analysing the empirical datasets. First, in the China sample, due to the availability of direct method disclosure of cash flow, the incremental power of cash flow disaggregation is also examined. The empirical results suggest that the direct disclosure of cash flow is really helpful in predicting future cash flows. The conclusion is consistent with previous studies such as Cheng and Hollie (2008) and Orpurt and Zang (2009), but this thesis is among the first to examine this argument based on out-of-sample comparison. In addition, the DKW assertion that using earnings could result in better prediction of future cash flow than using cash flow is not supported by all empirical datasets. In the U.K. listed sample, the random walk model outperforms the DKW model in the multi-period competition by both criteria.
Apart from the panel models, the Bayesian model and the grey-box model, it is noteworthy that the benchmark models, i.e. random walk model, theoretical DKW model and pooled regression BCN model, despite their simplicity, provide good performance difficult to surpass in many cases. For instance, for the China data, the random walk model has the lowest average rank in multi-period out-of-sample predictions. Therefore, the simple models are neither useless nor weak, especially when considering that the amount of calculation required is minimal.
142
Following the comparison of cash flow prediction models, the DSCF model is also applied for the listed firms of the three countries. The outputs of the model are theoretical stock values. A by-product of the DSCF model is the market RRA coefficient at a particular time. All the three datasets include the 2008 global financial crisis. During the financial crisis, the market indexes are closely related with the market RRA coefficients. The peaks of the market indexes correspond to a local minimum of the RRA coefficients and the fall of the market indexes is accompanied by an increasing RRA coefficient. Based on this implication, even though the stock markets of the three countries are all on obviously upward trends, it is perhaps more risky in U.K. and China stock markets than in U.S. market because the RRA coefficients of U.S. market are not on an obviously declining trend.
Through the DSCF model, cash flow prediction models could be examined for approximating the unobservable market expectation. The cash flow prediction model that results in theoretical stock values fitting the actual share prices better should be seen as a closer process to the market expectation of cash flow. It is not necessary that a better cash flow prediction model will better fit market expectation, but any exception will be a challenge towards efficient market hypothesis. The results of U.S. and U.K. markets suggest that a random walk model has a better fitness for the market expectation of future cash flows. This could be a good implication to investors because there will be great scope for strategy development, exploiting superior cash flow prediction models that may not be taken into account by the market. The theoretical value of a share can be used to determine a stock’s expected return and thus predict the price’s direction of movement. In the U.S. study, simple portfolios are constructed on conditions that are designed to select undervalued stocks. The portfolios have high returns compared with the contemporary market index. The portfolios are updated at a very low frequency and thus the return will be minimally impacted even if transaction costs are involved. When the superior power of the grey-box model in predicting cash flow is employed to design the investment strategy, the constructed portfolios could have even achieved a better performance. For the China data, an additional test has been conducted on the directions of price movement. The DSCF model has been applied to predict the directions of price movement, which is compared with the result of naive prediction, i.e. all shares will go up (or down). A two-sample t-test
143
suggests that the DSCF model could have superior power in making predictions for market prices.
In conclusion, the above comparisons of the empirical results by analysing four different datasets show that it is important to look into special features within each dataset, which might result in different modelling performances and conclusions. Cash flow behaviours of different countries have their own particular patterns, even though some similarities are shared. Even for firms within the same country, the cash flows may be captured by different models according to the firms’ operating conditions. It should also be noted that the samples used in this thesis have limitations. For instance, the U.K. unlisted firms only provide 10 years of observations, which restricts the extent to which they can be compared with their listed counterparts. Similarly, it should be noted that the disclosure behaviours of Chinese firms are different from those in U.K. and U.S. markets, mainly in the period of making disclosure. Chinese firms tend to disclose financial information during the first four months of year while U.K and U.S. firms’ disclosures span the whole year. As a result, the analysis of Chinese market for the rest 8 months starting from May becomes difficult and mostly uncertain as there are hardly any firms making disclosure during such periods. This is an obstacle of undertaking continuous study and analysis when the stock market is to be studied. It may require more assumptions to validate the study and make further analysis.
144