Curve Fitting Of Time Series With Equidistant Sampling Points

Time series is a data series which is tightly related to time sequence. Time series analysis is using statistical method to analysis this group of data set in order to find its pattern or to predict its future trend. Time series analysis is usually use in microeconomic control, systematic area development program, enterprise management and market potential prediction. Financial market usually appears as time series and it is a very important part in people's life, so to analysis financial time series has become an important research direction to economist and computer experts.Time series analysis needs continuous sample point to analysis its pattern more accurately, but in real life as sample points are always collect by observed data which are usually sparse, so they need curve fitting to fix the leak of sample points. This paper mainly discusses the specific example of this problem, curve fitting of time series with equidistant sampling points.Recently with the improvement of financial market, mutual fund has become more and more popular in population and takes an important part of investment field of our country. As the direct reveal of changes in fund asset, fund net value indicates the business circumstance of Fund Companies. In our country, mutual fund is divided into two kinds, the open fund and closed fund, and this paper is concentrated on the second kind. But its net value is announced only on Friday, it is important to predict the precise movement of fund net value in the rest days and that will be an important reference for mutual fund investors. This fund net value estimation can be seen as an real life application of the theoretical problems mentioned above.The mainly method for fund net value prediction can be divided into two kinds, one is using economic method to analyze financial report of Fund Companies, another way is technical method which makes use of statistic and artificial intelligent theories to find the price trend. This paper combined both of the above two methods, firstly dig out all features which may influence the fund net value using text analyst process, then predict it with statistic regression, and finally set several profile standards to determine the accuracy of prediction. Following parts are mainly content of this paper:1) Abstract the fund net value prediction problem to the theoretical question curve fitting of time series with equidistant sampling point, in order to interpolate sparsely sampling points into continuous ones which becomes training samples for fund net value prediction model in the following;2) Supporting vector regression is used in this paper to train certain features which has been extract from text analyst, and during the training process a lot of parameter adjustment has been done. And residual analysis is involved to mix up for errors which was included during experiment period, this amends the result of current model make the best practice of this problem with SVR model;3) This paper use the option entropy pricing theory which shows the fluctuation of fund net value can be simulated by the combination of the volatility of fund invested product price. So we can simplified certain market circumstance and reference the maximum entropy theory to come up with the stock investment combination invariance model (ME1) and industry investment invariance model (ME2), and then summarize a series of formula to practice in engineering;4) In order to completely compare of fund net value predict results, we include six evaluating standards, these standards evaluate the fitness of predict net value comparing with actual net value which gives investors more judgment rules and gives more evidence for model improvement. Experiment of this paper is based on the data center and data text analysis model of Haitian Yuan knowledge service platform. In the comparison of SVR model and ME model we finally choose ME1 model due to its perfect performance and convenience. Results of fund net value prediction is now displaying in Haitian Yuan closed fund website and all the data are continuance adjustment by model improvement, the precision of final result has already achieve the leading level of this field and has gain extensive attention among investors.