Quantitative investment traditional index investment decision vs Monte Carlo simulation method

Preface

Management scale about 5000 Billion 、 Yield Once defeated buffett Of Renaissance fund , stay 2020 year 11 In June, it was revealed that its three products 10 month losses 20%！

For beginners of quantitative investment , What should we value Traditional indicators , Or something else Probability and statistics , or machine learning And more advanced methods ？

disclaimer

Nothing in this vision and analysis should be interpreted as investment advice , Past performance does not necessarily indicate future results .

---

What is the prediction effect of Monte Carlo simulation method on stocks ？

The simulated stock is NYSE:GM,2020 year 1 month 1 Japan ~2020 year 12 month 24 Daily line data of the day

When we choose 5: 5 Training set of : Prediction set when

It can be seen that the fitting effect on the training set is very good , Next, let's look at the performance on the test set

It can be seen that the trend of the test set is not well predicted . In order to further evaluate the effect of Monte Carlo simulation , We don't look at the actual price forecast , But just look at the direction , If the final price of the test range is greater than the final price of the training range , We define it as income , vice versa . When the predicted direction is consistent with the actual direction , We think the forecast is accurate

And then we do it every 100 Statistical iterations 100~1000 The forecast of this time is shown in the figure above , It is found that the random effect is Failure Is far greater than Success The situation of .

When we choose 9: 1 Training set of : Prediction set when

You may feel that 5: 5 Training set of : The prediction set is not reliable , Maybe because the prediction set is too long . Then we'll have more training sets , The forecast set is smaller , Let's see if the situation is as expected

The first is to show the situation on the training set , Because the training set is too long , The fitting effect is slightly poor

See the performance on the prediction set , It is not as good as we think , On the contrary, the opposite trend has occurred

every other 100 Statistical iterations 100~1000 The forecast of this time is shown in the figure above , Compared with the previous forecast, the situation has improved to some extent , But Monte Carlo simulation is a model that uses random numbers , Is the result really good ？ You know, tossing a coin to judge whether it goes up or down is 50% The probability prediction is correct , However, the performance of Monte Carlo shows great uncertainty , For this result , Simple Monte Carlo simulation is not desirable .

Does Monte Carlo simulation really play such a role in the stock market ？ This is really a good research topic . Do you need more data and more comprehensive analysis ？ Consider the correlation with other stock prices ？ Add correction model ？

---

Traditional index decision-making

So what is the effect of using traditional indicators to make decisions ？ Or adopt 2020 year 1 month 1 Japan ~2020 year 12 month 24 Japanese NYSE:GM data

use BOLL and MACD Double index decision

First, put the trading effect picture directly , Each time is the total amount of purchase and sale , Regardless of the handling fee . According to the color habit of American stocks , Red in the candle chart is the opening price > Closing price , Green is the closing price > Opening price . Then the left side of the large column represents the buying point 、 The right side indicates the selling point , The red column is a losing trade , The green column is a profitable transaction

The situation of each transaction and the final yield

All indicators considered are plotted as follows

Observe the back test results and find , Our decision to stop loss in time avoids 2020 year 3 The sharp fall in January , And seize the opportunity in the callback , Success makes a profit in the callback process 36.24%, Much higher than the stock in 2020 Year's increase .

In addition to A Stocks also have good performance , Last week, Baijiu stocks fell into the callback trend , The strategy did not choose to stop profit and sell Baijiu stocks , But choose to keep holding , As things stand , The choice is relatively correct .
With “000799.SZ Alcoholic liquor ” For example , The rest of the stocks can try for themselves

disclaimer ： Nothing in this vision and analysis should be interpreted as investment advice , Past performance does not necessarily indicate future results .

---

Code implementation of traditional strategy

For the above mentioned BOLL and MACD Combined policy code is too long , You can view it in GitHub Read and download https://github.com/ExileSaber/Quantify-Investments

In addition, for this strategy A Stocks also performed well , But this is not investment advice , Just introduce relevant strategies , And it should be noted that the back test results do not represent the effect of the firm offer

---

summary

Of course, the back test results do not mean that there will be a good performance in the future firm trading , But compared with Monte Carlo simulation, which has poor explanation , The traditional index decision-making is much better in explanation , Every point has a definition of buy and sell . What kind of model should be used 、 Strategy to do a good job in analysis depends on the investors' own decision

disclaimer ： Nothing in this vision and analysis should be interpreted as investment advice , Past performance does not necessarily indicate future results .