Méthodes de prévision sur les marchés financiers: Une recherche sur la prévision des marchés financiers et le ratio différence des taux d'obligations en Turquie

As the starting question of the thesis, instead of traditional financial ratios such as Price/Earnings and Dividend Yield, it has been determined whether the information hidden in the relationship between bond and stock returns can be used to predict movements in the capital markets. Although the relevant financial data do not go back very far, there are different indices with different methodologies, especially for bond yields, and these contain various problems within themselves. The analysed data have not been used very much in past literature. In this respect, the data to be used before the study will be examined in depth. This enabled the study to reveal its strengths and weaknesses by compiling in detail the data that could be the subject of the study for Turkey. With the research, it is explained which of the old BIST indices, BIST-KYD indices, TEB BENCH Index or IECM2Y Indices calculated by Bloomberg are the most suitable benchmark (more explanatory) for bond yields in Turkey. According to the results of open discussions with portfolio management companies on this subject, a dominant opinion has emerged that the IECM2Y index, which is better known in the market and recommended to be used by public authorities, will also be more suitable for international use. For share returns, Berge & Ziemba (2003) stated that finance variables such as E/P (the inverse of Price Earnings ratio, expressed as return) and dividend yield were intensively examined to find the predictive power of stock market returns. P/E ratio is used for share side return. In this area, the daily P/E ratios of the BIST 100 Index, which is used as an indicator index by Borsa İstanbul, announced by Borsa İstanbul, Reuters and Bloomberg, were obtained from Bloomberg. During the research period, it was extended to the oldest date that could be extended. This date is 21 June 2006, which is the start date of the IECM2Y Index, which shows the 2-year compound interest rates. It started from this date and was brought until 21 June 2018, which is 12 years later. It is quite sufficient for working with more than 10 years in total period and more than 3000 pieces of data at consideration. The Bond-Stock Earnings Yield Differential model (BSEYD) used in the study was first proposed by Berge and Ziemba in 2003. The model was reviewed by Lleo and Ziemba in 2013 as to whether it has the ability to predict crises better than the Campbell and Shiller model. Published as a discussion article by The London School of Economics and Political Science (LSE) in September 2014. The model was searched from YÖK Thesis database, Google Scholar and Science Direct and it was determined that it was not used in any study on Turkey. In this respect, we can say that it is an original study for Turkey. In addition, while researching the bond indices to be used, the bond indices calculated and being calculated in Turkey are explained in chronological order and serve as a guide for researchers who will do research in this field. The relevant ratios in the current literature were examined in terms of Turkey. These are the Long-Term Bond-Share Yield Ratio (GEYR, Gilt-Equity Yield Ratio), the Bond-to-Share Yield Ratio (BEYR, Bond Equity Yield Ratio) and the Bond-Share Spread Ratio (BSEYD, Bond-Stock Earnings Yield Differential) is the rate. BSEYD ratio was used in our study. This ratio was examined with the Markov Switching Model. The Markov Regime Change Model was evaluated through the econometric software program called OxMetrics. The results are as follows. During the examined period (21/06/2006 – 21/06/2018), the model gave signals 8 times. It is possible to read the general picture from Figure 4.3 below. The portfolio, which started with bonds, went between bonds and shares, then turned into shares as of the end of 2009 and continued in the denominator for a long time until November 2017. It has been converted into bonds as of November 2, 2017. If we try to explain the accuracy of the signal with the current economic situation other than the calculation, it is possible to say that it works correctly. Because the BIST 100 index was around 110,000 points, which is a record level, at the time the signal was working, and therefore the price/earnings ratios were up (earnings/price ratios were down). At the same time, bond yields have risen. Two factors increase the BSEYD value that we used in the model. The model last signalled with the value of 07 March 2018. With this signal, it means that the yield of the bond has increased compared to the share, or in other words, the prices of the shares are higher (expensive) compared to the bond. As of the last date, it still holds the portfolio in bonds. While the BSEYD rate was higher until 2009, it has taken lower values since 2009. Recently, it has started to take high values again, but we cannot say that it is a different pattern yet. In fact, before and after the date of 11 June 2008, when it first signalled, can be examined as two separate periods. Because they stand as two very different patterns. The reason for this difference may be the subject of another study. Possible reasons may be high inflation, high risk perception in Turkey and very high bond yields. Based on these two different patterns, a second model was established and, in this model, the second period was examined within itself. The model is exactly the same, only the period in which it is applied is different. With this difference, first of all, the two different periods will be examined separately and at the same time, since the second period is a time period within the first period, it will be seen whether signals are produced for different dates in the common time period. As a starting point for the second period, the period from the last quarter of 2012, when Turkey was given the investment grade rating for the first time in history, until November 2017, when the first signal was produced after holding the denominator for a long time in the 1st Model. When we examine Model 1 mentioned in the introduction in detail, it is seen that the model gives signals 8 times during the period. The study will start from 21 June 2006, the start date of the IECM2Y Index, which shows the 2-year compound bond rates, and will be brought to 21 June 2018, which is 12 years later. The portfolio started with bonds. With the 5. signal, it turned into a numerator at the end of 2009 and continued as a denominator for a long time until November 2017. It turned into a bond as of March 7, 2018 and continued in the bond, which did not produce a new signal until the last updated data. As of the last situation, it can be said that the bond prices are cheap compared to the share prices or that the share prices are more expensive than the bond prices. At the same time, a correction in share prices can be expected. If we try to explain the accuracy of the signal verbally with the current economic situation, we can say that it works correctly. In November 2017, when the signal was working, the BIST 100 index was around 110,000 points, which was a record level, and thus the price/earnings ratios were up (earnings/price ratios were down). At the same time, bond yields are higher than their long-term averages. Both factors increase the BSEYD value that we used in the model. The model last signalled with the value of 06 March 2018. With this signal, it means that the yield of the bond has increased compared to the share, or in other words, the prices of the shares are higher (expensive) compared to the bond. As of the last period, he still holds the portfolio in bonds. When we examine Model 1 in detail, which was considered at the beginning and mentioned above, it can be said that it consists of two different periods. While the BSEYD rate was higher until 2009, it has taken lower values since 2009. In fact, before and after the date of 06 November 2008, when it first signalled, can be examined as two separate periods. Because they stand as two very different patterns. The reason for this difference may be the subject of another study. Possible reasons may be high inflation, high risk perception in Turkey and very high bond yields. Based on these two different patterns, a second model was established and, in this model, the second period was examined within itself. The model is exactly the same, only the period in which it is applied is different. With this difference, first of all, the two different periods will be examined separately and at the same time, since the second period is a time period within the first period, it will be seen whether signals are produced for different dates in the common time period. As a starting point for the second period, the last quarter of 2012, when Turkey was awarded the investment grade for the first time in history, will be taken. The dates between 13/09/2012 and 28/11/2017 were looked at. It is seen that the model gives signals 7 times during the examined period. Although it had a shorter period than Model 1, it gave a large number of signals. Here, it is possible to say that this result emerged because it is easier to deviate from the mean value in the short term compared to the long term. In addition, since more signals will mean more returns, it would be appropriate to work with more up-to-date data instead of looking at very long periods in using the model. One thing to be aware of is whether there is a seasonal variation in the graph. However, since looking at the model for very short periods will give too many buy/sell signals, it will deviate from the aim of hedging (protecting) the initial long-term portfolio and obtaining better periodic returns. When we examine Model 2, the portfolio started with the share. In Model 1, this period portfolio is in the denominator. With the 5th signal in Model 1, the portfolio turned into a share at the beginning of 2010 and continued as a denominator for a long time until November 2017. However, in Model 2, signals were received from time to time and the portfolio changed between stocks and bonds. Model 1 turned into a bond as of November 2, 2017 and continued in the bond, which did not produce a new signal as of January 10, 2017 and continued in the bond, which did not produce a new signal until the last updated data. In this respect, we can say that Model 2, which looks at a shorter and current period, produces signals earlier. As of the last situation, it is possible to say that bond prices are cheaper than share prices or vice versa. At the same time, a correction in share prices can be expected. The most appropriate model for Turkey was tried to be developed and the question "Can bond-stock earnings yield differential model be an estimation tool in the Turkish financial market?" The answer to the question has been sought. It has been concluded that the bond-share yield ratio can be used as an estimation tool in the 12- year period and the selected shorter sub-period. The model makes a choice between the signals it produces and the bond and stock, which are seen as two rival assets. It suggests a transition to another by signalling a switch among bond and stock according to the change in returns. It recommends exiting the share market, especially in case of collapses in the share market. Therefore, the decline in the stock market is avoided while benefiting from the bond market returns. As an investment strategy, it will warn both individual and institutional investors. Funds can be issued on a strategy index to be produced using the model. For long term investment options, the model is highly useful considering bear markets at stock.