Trading Strategy In High Frequency Financial Data And Volatility Analysis

The purpose of running a financial market is to contribute to the regression of asset price to its value. Since various parts in and out of this market show their opinions with bids and asks, which lead to the chaos of prices. Among chaotic prices, professionals cultivate returns by predicting trends and estimating volatilities. And no capital asset is required in this business. Such a free lunch stimulates strong speculation in financial markets. From 21th century on, lots of trading strategy-based investment prevailed over financial market. This kind of investments is quite different from the definition of CAPM theory, it generates trading signals by propitiate technical indicators, so called technical analysis. This strategy has burden heavy critics from academia for decades. However technical analysis survived and even shew prosperity in the late 1990s. In this research, two reasons of many are given to explain such a prosperity.On the one side, the famous EMH theory born in 1970s had been doubted for years, and hundreds of empirical analysis shew the fact that efficient market lives only in mathematical models. In the real world, investor is neither fully-informed nor homogeneous rationale. Two concrete discoveries from behavior finance are moment trading and herding behavior. The former tells us that investors who are lack of rationale, tend to follow the price and buy when price goes up, and sell when down. While the latter indicates that the irrational behavior at market level will force price divert from its value for quite a while, and hinder probable recursive process in a longer period. So in a market existing strong herding behaviors, the function of asset allocation is quite trivial.On the other side, discussions against technical analysis lead to the generation of resampling methods, sliding estimation, statistical analysis-embedded technical analysis, bayesian judgement for choosing trading indicators, data selection techniques, and etc. However, the mainstream opinion still goes like "given the condition of transaction cost, technical analysis is inefficient" . Then following question appears hereby: if technical analysis is useless, then why it still lives with us and quite well? Is it because of defaults hidden in the opponents' critics, or simply because of the stickiness of professionals and analysts? After a broad review, we find aspects which had been neglected by mainstream study but quite essential to the success of technical analysis: (1) most recent empirical research are based on filter rules, moving averages and etc, which are easy for programming and quite popular among analysts. But there are still other hundreds of technical indicators being used in the same time. So by testing the defaults of the few and alleging all are not suitable and not reasonable as well. (2) technical indicator aims to capture variation in prices, not to explain it. In order to do well in predicting variations, more inferior predictions can be made. The extremes are never the favor of statistical tests, and a great number of insignificant trading signals can also lower the significant level. (3) the timing ability are essential to technical analysis, while it rarely appears on the formal research works. Among those empirical works reviewed, daily close price are mostly used, which obviously only supplies the result of daily trading, daily prices variations, which is the main source of interest for technical analysis, can be found in close price, plus some even use weekly data, whose conclusions maybe far from reality.This study use a huge by-hand dataset over 1.3 gigabytes, containing each transaction records of 40 stocks in a year. Not a dataset used in all research works we reviewed, no matter in quality or in data scale, can be compared with ours. After one month's empirical study, results and conclusions show values for theoretical references, and it is quite valuable for practical designing of micro-level trading strategy as well. The main works done can be concluded as follows.Firstly, we use genetic algorithm to give joint technical indicators a better estimation for their parameters. Some basic characteristics of trading strategy are found: (1) trading strategy is quite sensitive to the price variation,the greater the variation, the higher the gains; otherwise less or few. (2) trading strategy can be sensitive to the price sampling intervals. Given the same strategy, the different sampling frequency can generate many different results. (3) due to its parallel trading property, the behavior of its returns cannot be fitted in the traditional statistical analysis. Yet the value at risk method shows some potential to solve this problem.Secondly, we use classical pairs trading to select pairing assets, which is quite easy and has good property of parallel trading. Strong price-driven mechanism and parallel trading, can give huge return and lost together. This reminds us that keeping an eye on liquidity and risk management is the key to success of trading strategy. Further considering the general attributes of financial times series, co-integration method is introduced in this study. After some modification, i.e., using the difference of logged prices instead of logged prices, we show that co-integration based pairs trading method outperforms classical one in many ways. Especially, in the sampling frequency of 30 and 60 minutes, even considering transaction costs, co-integration pairs trading does the same works as the performance of considering no transaction costs, and their VaRs are also close. So we concludes that co-integration methods shows better fitness for dynamical stability for pairs assets.Thirdly, under the arbitrage pricing theory(APT), we use the difference of returns to build a state space model. To make better prediction, kalman filter is incorporated. Results show that strategy-based portfolio's daily returns has extremely high observations, and positive skewness shows the potential of positive gains. Facing with 20 times lost, its 40 times gain is still lucrative. Again here we see, extremes represents eagerness of pursuing extreme profits.Lastly, to give a more general view of these trading strategies discussed, we incorporate ideas of market neutral investing strategy and APT theory to decompose sources of returns of trading strategy. After an in-depth discussion about market neutral and risk premium, we conclude that: market neutral strategy is simply arbitrage over risks which is treated as unique risk or has not been considered by most investors. Additionally, we develop a easy and clear dynamic model to explain the mechanism that how strategy impacts market volatility. This model shows clearly that profit-pursuing investment-swinging behavior is one main source of volatility in financial market.