But it must be established that quantitative investment is not the same as high-frequency trading. If the services are divided by frequency, they can be divided into: high frequency: ticke nanosecond 1s, medium and low frequency: 1s~2h, ultra-low frequency: 1d~2w, etc. Long-term investment in high-frequency trading requires very high delay time, characteristics and reliability, and requires a lot of hardware configuration costs and labor costs.
However, the requirements for hardware configuration of medium and low frequency transactions will be much lower. The key difference between myself and the fund management company lies in the optimization algorithm, and ordinary programs also have the ability to capture the trading data signal of this frequency. Don't talk nonsense, just say a word and do it at once! If we want to analyze A shares, or BTC, we must establish a set of natural environment by ourselves.
Generally, to build a quantitative platform, this process is necessary: setting up an account > building development tools > preparing data and materials in advance > developing and designing quantitative trading strategies > regression testing > simulated trading > real trading. 1. Set account (ignored here).
First of all, the development tools for building two popular service platforms at this stage are python and R language. These two languages express the framework of backtesting and the library of time series analysis and data analysis (both c++ and java are acceptable, but the threshold is relatively high).
Python: At this stage, it should be the first choice for quantitative analysis, because the related open source frameworks are very colorful. R: Advanced optimization algorithm is more convenient and the community is more active. I chose Python, and the common backtesting architecture uses ZipLine and BackTrader.
Second, data information Prepare China's stock data in advance, which is given by some service providers, such as Tonglian Data and tushare；; You can get the data information of overseas securities from these data information, and then you can query them in the database. About the choice of database query, Mysql is generally used. If the amount of information is large (> 100G), mogodb can be used. It is not easy for me to have such a large amount of information.
Third, the development and design of quantitative trading strategy When it comes to trading optimization algorithms, we usually think of powerful AI words such as deep learning, Markov entity model, data analysis, deep neural network and neural network. But ordinary gamers basically don't need it.
Ordinary investors can adopt simple and efficient optimization algorithms:
1. Programmatic trading of your own actual operations and ideas, such as: Sanlianyang, buying low-priced stocks or what magical practical skills you have heard, are all completed by coding, and then back-tested by using historical records.
2. Traditional index value trading: moving average, MACD, bollinger band, etc. Basic theory of candle diagram, RSI, Gann theory. This pure technical index analysis index value is only effective under special circumstances. Everyone has heard of the turtle trading law, and probably all think it makes sense. But what is the specific situation? If we use A-share or foreign exchange data to test, we will find that the long-term yield is not particularly good.
3. Multi-factor stock selection: Every investor has his own basic theory of stock selection. For example, some people will look at the price-to-book ratio, stock turnover rate, price-to-book ratio, market position and trading volume. This selection factor is very simple, but if you choose from thousands of stocks, you usually need a lot of energy. Program flow can deal with this kind of problem particularly well. If you are an advanced player, you can also try the advanced optimization algorithm.
Such as deep learning and data analysis. The application of Internet big data in the financial investment industry is still in a gradual process. Judging from the information content at this stage, the yields of Internet big data funds are very good, such as the Baifa Index Fund cooperated by Baidu Search and GF Securities, and the Internet big data stock fund cooperated by Tencent and Harvest.
Fourth, the regression test If the actual effect of the back test is very good, and the indicators such as the rate of return, the maximum withdrawal rate and the Sharp value are all within the acceptable range, you will undoubtedly be excited, eager to buy and sell, and even gradually plan to establish a private investment fund. However, don't worry, it is best to simulate trading.
5. simulate trading. But before the firm offer, the simulated transaction must be done for one or two months. Many countermeasures with very good backtesting results may not perform well mainly in simulated trading situations. The history is fixed, and the main parameters can be constantly adjusted according to the back test, which makes the values of various indexes tend to extremes and sometimes leads to over-fitting of the optimization algorithm. Because the sales market has been varied, too rigid optimization algorithm can not be integrated into the transformation of the sales market.
The final practical effect of simulated trading generally lies in whether your program flow is flexible or not, and whether it is an excellent risk and fund allocation optimization algorithm.
Summary: As to whether it is reliable for me to make quantitative investment, the above steps have shown the practical and planned scheme, and the reliability is obvious. Whether I can make money depends on the training I received for a better place.