WO2021003791A1

WO2021003791A1 - Black swan event decision method and apparatus based on subjective and objective joint prediction

Info

Publication number: WO2021003791A1
Application number: PCT/CN2019/099406
Authority: WO
Inventors: 樊文晙; 都志辉
Original assignee: 清华大学
Priority date: 2019-07-11
Filing date: 2019-08-06
Publication date: 2021-01-14
Also published as: CN110458328A

Abstract

A black swan event decision method and apparatus based on subjective and objective joint prediction. The method comprises: obtaining a prediction event and a historical data set (S101); predicting the prediction event by means of a prediction model to obtain first prediction data, wherein the prediction model is generated by performing statistical analysis of the historical data set by means of a preset machine learning method (S102); searching a preset subjective experience database for second prediction data matching the prediction event (S103); and comparing the first prediction data and the second prediction data for difference processing to obtain a prediction difference value, and deciding the prediction event according to the prediction difference value (S104). By combining two different types of prediction methods, i.e., subjective and objective prediction methods, the method provides an effective black swan event decision index, which can be used for risk control, thereby greatly reducing losses.

Description

Black swan event decision-making method and device based on subjective and objective joint prediction

Cross references to related applications

This application claims the priority of the Chinese patent application number "201910623881.5" submitted by Tsinghua University on July 11, 2019 with the title of "Black Swan Event Decision Method and Device Based on Subjective and Objective Joint Prediction".

Technical field

The feedback of this application belongs to the technical field of risk control, and particularly relates to a black swan event decision-making method and device based on subjective and objective joint prediction.

Background technique

For risk management, the Black Swan Event (BSE), that is, an unexpected event that can cause great impact and consequences, often impacts the normal production order of nature and society. The black swan event is usually the opposite of the normal event. The two cannot happen at the same time. The normal event often has a greater probability of occurrence, and the black swan event has a smaller probability of occurrence. Therefore, people cannot treat all the negative and low-probability events of normal events as black swan events that will happen to prevent and deal with, because the cost is too great. Therefore, the challenge of the black swan event comes from the fact that it will have a huge impact once it occurs. It is too costly to focus on it, and it is very uncertain whether it will happen.

Summary of the invention

The feedback of this application aims to solve one of the technical problems in related technologies at least to a certain extent.

For this reason, one purpose of the feedback of this application is to propose a black swan event decision-making method based on subjective and objective joint prediction. This method combines subjective and objective prediction methods of two different types to give an effective black swan event decision-making. Indicators can be used for risk control and greatly reduce losses.

Another purpose of the feedback of this application is to propose a black swan event decision-making device based on subjective and objective joint prediction.

In order to achieve the above-mentioned purpose, the present application feedbacks that one embodiment proposes a black swan event decision-making method based on subjective and objective joint prediction, which includes:

Obtain predicted events and historical data sets;

Predicting the predicted event through a predictive model to obtain first predictive data; wherein the predictive model is generated by performing statistical analysis on the historical data set through a preset machine learning method;

Searching for second predicted data matching the predicted event in a preset subjective experience database;

The first prediction data and the second prediction data are compared for difference processing to obtain a prediction difference, and a decision is made on the prediction event according to the prediction difference.

The black swan event decision-making method based on subjective and objective joint prediction of the feedback embodiment of the present application obtains predicted events and historical data sets; predicts the predicted events through a prediction model to obtain the first predicted data; searches and searches in a preset subjective experience database The second predicted data matched by the predicted event; the first predicted data and the second predicted data are compared for difference processing to obtain the predicted difference, and the predicted event is decided according to the predicted difference. As a result, the reliability of predicted events can be predicted through subjective and objective prediction to improve the performance of identifying black swan events and risk control.

In addition, the black swan event decision-making method based on the subjective and objective joint prediction of the foregoing embodiment fed back according to the present application may also have the following additional technical features:

Further, in an embodiment of the feedback of the present application, the making a decision on the predicted event according to the predicted difference includes:

Determining that the predicted difference is less than a preset threshold, then determining that the predicted event is a security event;

If it is determined that the predicted difference is greater than or equal to the preset threshold, it is determined that the predicted event is a risk event.

Further, in an embodiment of feedback in the present application, the historical data set is historical data related to the predicted event within a preset period.

Further, in an embodiment of the feedback of this application, the historical data set is statistically analyzed through the logistic regression model of scikit-learn.

Further, in an embodiment of the feedback of this application, it further includes:

Acquiring a plurality of prediction data related to the prediction event;

Analyzing the plurality of prediction data related to the prediction event to generate the preset subjective experience database.

In order to achieve the above-mentioned purpose, another embodiment of the present application provides a black swan event decision-making device based on subjective and objective joint prediction, including:

The first acquisition module is used to acquire predicted events and historical data sets;

The objective decision-making module is used to predict the predicted event through a predictive model to obtain first predictive data; wherein the predictive model is generated by statistical analysis of the historical data set through a preset machine learning method;

The subjective decision-making module is used to search for second prediction data matching the predicted event in a preset subjective experience database;

The decision output module is configured to compare the first prediction data and the second prediction data for difference processing to obtain a prediction difference, and make a decision on the prediction event according to the prediction difference.

The black swan event decision-making device based on subjective and objective joint prediction in the feedback embodiment of the present application obtains predicted events and historical data sets; predicts the predicted events through a prediction model to obtain the first predicted data; searches and searches in a preset subjective experience database The second predicted data matched by the predicted event; the first predicted data and the second predicted data are compared for difference processing to obtain the predicted difference, and the predicted event is decided according to the predicted difference. As a result, the reliability of predicted events can be predicted through subjective and objective prediction to improve the performance of identifying black swan events and risk control.

In addition, the black swan event decision-making device based on the subjective and objective joint prediction of the foregoing embodiment fed back according to the present application may also have the following additional technical features:

The second obtaining module is used to obtain a plurality of prediction data related to the prediction event;

The generating module is configured to analyze the plurality of predicted data related to the predicted event to generate the preset subjective experience database.

The additional aspects and advantages of the feedback of this application will be partly given in the following description, and some will become obvious from the following description, or be learned through the practice of feedback of this application.

Description of the drawings

The above-mentioned and/or additional aspects and advantages fed back by this application will become obvious and easy to understand from the following description of the embodiments in conjunction with the accompanying drawings, in which:

Figure 1 is a flow chart of a black swan event decision-making method based on subjective and objective joint prediction according to an embodiment of the present application;

FIG. 2 is a schematic diagram of evaluating 48 group matches of the 2018 World Cup according to an embodiment of the present application;

Fig. 3 is a schematic structural diagram of a black swan event decision device based on subjective and objective joint prediction according to an embodiment of the present application.

Detailed ways

The following describes the feedback embodiments of the present application in detail. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals indicate the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the drawings are exemplary, and are intended to explain the feedback of this application, and should not be understood as a limitation on the feedback of this application.

The following describes the black swan event decision-making method and device based on subjective and objective joint prediction proposed according to the feedback embodiment of the present application with reference to the accompanying drawings.

First, the black swan event decision method based on subjective and objective joint prediction proposed according to the feedback embodiment of the present application will be described with reference to the accompanying drawings.

The key problem to be solved by the feedback of this application is to provide an indicator that can classify which small probability events are indeed impossible to occur, but some small probability events are very likely to occur and become black swan events. The feedback method of this application needs to combine subjective and objective predictions of target events. Subjective Prediction (SP) refers to the judgment made by the human brain's reflection on the possibility of an event. In this prediction and judgment, the person's own knowledge, experience, and intuition play a major role. In addition, human judgment will pay more attention to the current real-time events and changes that may have an impact on the occurrence of the target event. On the contrary, Objective Prediction (OP) refers to the probability that an event may occur through statistics on historical data. That is, the probability of this event occurring in the past is used to predict the probability that it will occur in the future. This method itself has a good predictive effect on high-probability events, because events that have occurred in history are likely to happen again. But this method is also easy to misjudge the black swan event, because the black swan event is usually an event that has never occurred or rarely occurred in history. The specific process of this method is introduced below.

FIG. 1 is a flowchart of a black swan event decision-making method based on subjective and objective joint prediction according to an embodiment of the present application.

As shown in Figure 1, the black swan event decision-making method based on subjective and objective joint prediction includes the following steps:

In step S101, a predicted event and a historical data set are obtained.

Further, the historical data set is historical data related to the predicted event in a preset period.

Specifically, the historical data set is the historical data related to the predicted event in a period of time. For example, to predict the outcome of team A and team B in a game, you can obtain information about team A during the ten years. The outcome of the match against team B. It is understandable that the preset period should be set according to the actual situation, and the preset period should be large enough.

In step S102, the prediction event is predicted by the prediction model to obtain the first prediction data; wherein the preset model is generated by statistical analysis of the historical data set through a preset machine learning method.

Specifically, the prediction model is generated by statistical analysis of the historical data set related to the predicted event through a preset machine learning method. The preset machine learning method can be multiple, and the prediction model is obtained by learning the historical data set. , In order to achieve the prediction of the predicted event to obtain the first predicted data.

It is understandable that the first prediction data is data obtained through objective prediction.

In step S103, search for second predicted data matching the predicted event in a preset subjective experience database.

Furthermore, there are many ways to predict the predicted event through subjective prediction. As a possible way, a subjective experience database is established, and the prediction is made by searching for predicted data matching the predicted event in the subjective experience database, including: :

Obtain multiple forecast data related to forecast events;

Analyze multiple predicted data related to predicted events to generate a preset subjective experience database.

It is understandable that when predicting a predicted event, the objective prediction result can be obtained through the analysis of machine learning methods, or the subjective prediction result can be obtained through human subjective experience. In subjective prediction, By obtaining subjective data related to predicted events, such as through online or offline surveys, collecting people’s subjective predicted data on predicted events, and then processing these data, a preset subjective prediction database is generated through the database generation method. When predicting a certain predicted event, the predicted data matching the predicted event is obtained by searching in the subjective prediction database.

In step S104, the first prediction data and the second prediction data are compared for difference processing to obtain a prediction difference, and the prediction event is decided according to the prediction difference.

Further, making decisions on forecast events based on forecast differences includes:

If it is judged that the predicted difference is less than the preset threshold, the predicted event is determined to be a security event;

It is determined that the predicted difference is greater than or equal to the preset threshold, and the predicted event is determined to be a risk event.

Specifically, the prediction data is obtained through objective prediction and subjective prediction respectively, and the prediction difference between the two is processed to obtain the prediction difference, and the prediction difference is compared with the preset threshold. If the prediction difference is less than the preset threshold, then It shows that subjective prediction and objective prediction are basically the same, so this consistent prediction judgment is used, which points out that a high probability event will occur, or a small probability event will not occur. If the prediction difference is greater than the preset threshold, it means that the subjective prediction is different from the objective prediction, and the difference is large. You can boldly suspect that there will be a black swan event, that is, a high probability event does not occur or a small probability event occurs instead.

It is understandable that the decision index can usually be expressed as a threshold. If the difference between the subjective and objective predictions exceeds this threshold, the black swan event is considered to occur, and if the threshold is lower than the threshold, the two judgments are considered consistent and the accident will not occur.

In summary, on the one hand, machine learning methods are used to make predictions by statistical analysis of event history data, and on the other hand, human experience and observations are used to make predictions. Then compare the two predicted data and use the difference as a decision indicator. When the decision index is lower than a certain threshold, the original prediction result is credible. And when the decision index is higher than this threshold, the original prediction result is not credible, and its reverse event, namely the black swan event, has a great probability of occurring. By combining the subjective and objective prediction methods of two different types, an effective black swan event decision-making index is given, which can be used for risk control and greatly reduce losses.

The feedback method of this application will be described in detail below through a specific embodiment.

Football is the world's largest sport. The four-year World Cup football match is the highest level of football competition and has a lot of attention. In addition to the basic factor of the difference in strength between the two sides, there are many off-field factors that affect the outcome of a World Cup match. Therefore, there are often unpopular games, which is the so-called black swan incident. The following will use the 2018 Russia World Cup as an example to illustrate a specific application of feedback in this application.

As we all know, there are two teams A and B participating in a game, and the result of the game is nothing more than A win, B win, or draw. Each result will have a certain probability of occurrence, and the sum of the probability of occurrence of the three is 1. Bookmakers (such as Bet356) will open handicap (Odds Handicap) before the game. Fans will place bets on the game. The result of the betting can reflect people's predictions on the outcome of the game. Turn this prediction into a subjective prediction of a football game. Assuming that team A wins the odds of a, the probability of B wins is b, and the draw is c. Generally, betting will be stopped 20 minutes before the game, and the odds will be stable. You can use formulas (1), (2) , (3) to calculate the probability of team A winning (P _a-SP ), the probability of winning B (P _b-SP ) and the probability of a tie (P _c-SP ):

P _a-SP =(1/a)/((1/a)+(1/b)+(1/c)) (1)

P _b-SP =(1/b)/((1/a)+(1/b)+(1/c)) (2)

P _c-SP =(1/c)/((1/a)+(1/b)+(1/c)) (3)

On the other hand, historical data can be trained by machine learning models to make objective predictions. The probabilities of the three outcomes of a game are recorded as P _a-OP , P _b-OP and P _c-SP . In this example, scikit-learn's Logistic Regression model is used to train and classify the data. FIFA has been conducting international rankings of national teams since 1930, so it is necessary to obtain the results of all competitions that have occurred between various national teams since 1930 as a training history data set. In this data set, select the matches that the national teams that participated in the 2018 World Cup have participated in (a game as long as one side is the 2018 World Cup team). The data only focuses on the values of three characteristics, the home team, the away team and the result of the game. Label the results of the game from the perspective of the home team, with 2 representing the home team winning, 1 representing a tie, and 0 representing the home team losing. Then, one-hot encoding is performed on the game records between all national teams after the above filtering with these three feature values, so that these data can be loaded and processed by the scikit-learn machine learning model. The preprocessed data set so far is the total set of training and testing. Use 70% of the records for training and 30% for testing. Then, one-hot coding is also performed on the matchups of the 32 teams participating in the 2018 World Cup in the group stage. Since there is no home or away team in the World Cup, here is the international ranking before the start of the 2018 World Cup to distinguish the popularity of the team in the World Cup. The most popular team is considered the home team in the match. The prediction set after this one-hot encoding also needs to be consistent with the previous one-hot encoding of the training and test set, so add the missing encoding column (denoted as 0).

According to the above method, before the start of each game, the result of subjective and objective prediction can be obtained. Use the largest difference between the subjective and objective predictions of the three game results as the decision making indicator (DMI), namely:

DMI=max(|P _a-SP -P _a-OP |,|P _b-SP -P _b-OP |,|P _c-SP -P _b-OP |) (4)

Here defines the unpopular game (black swan event) as: 1. The home team (ie the strong team, the one with high international ranking) has a 50% higher probability of winning than the away team (ie the weak team), but the result of the game is that the weak team wins. Or the game is tied; 2. The home team's winning probability exceeds the visiting team's winning probability by 20% and less than 50%, but the result is that the weak team wins. In addition, if the difference between the winning probability of a strong team and the winning probability of a weak team is less than 20%, and the two team instances are considered to be relatively close, then any game result is acceptable and there will be no upsets, so the game can be ignored (Ignored-Game ,IG), not included in the consideration of the upset competition.

For combined subjective and objective predictions, a game is regarded as a negligible game, and both subjective and objective predictions are required to consider it as a negligible game. Except for negligible games, all other games are considered risk games (RG). For a risky game, use judgment indicators based on subjective and objective joint prediction to predict whether it will become a black swan event: if DMI exceeds a certain threshold, for example, set to 5% in this example, it is considered that it will become black Swan event, otherwise the game is a safe game (Safe Game, SG).

The above-explained DMI-based decision-making method was used to evaluate the 48 group matches of the 2018 World Cup, as shown in Figure 2. The horizontal dashed line is the threshold value of 0.05, and those smaller than the threshold value are SGs for safe games.

Give an example of an unpopular game to illustrate. This unpopular match was Germany against Mexico. Before the start of the match, according to objective predictions, the probability of Germany winning was 0.589, the probability of a draw was 0.258, and the probability of Mexico winning was 0.153; the odds before the start of the game were locked at 1.5 for Germany to win and 4.5 for a draw. , Mexico wins 7.5, which can be converted into people’s subjective predictions. The probability of Germany winning is 0.652, the probability of a tie is 0.217, and the probability of Mexico winning is 0.130. So we can first calculate that this is an RG, and then find that its DMI value is 0.063, which exceeds the threshold of 0.05, so we boldly predict that the game will be upset. The result verified that the prediction was correct. The score on June 17th was that the German team lost 0:1 to Mexico.

Furthermore, the standard Accuracy, Precision, Recall, and F1 metrics are used to evaluate the subjective prediction, objective prediction, and subjective and objective joint prediction of the team's judgment effect of the unpopular match in the group stage, see Table 1. It can be seen that the combined subjective and objective forecasts outperform the other two forecasts in various measurement standards.

Table 1 Forecast results of different forecasting methods

To	SOPSOP	SPSP	OPOP
AccuracyAccuracy	0.5830.583	0.43750.4375	0.4160.416
PrecisionPrecision	0.3540.354	0.250.25	0.2430.243
RecallRecall	11	11	11
F1F1	0.5230.523	0.40.4	0.3910.391

In addition, an investment strategy can also be given based on the above method. Assuming that 48 principals can be invested in 48 games, the investment strategy is:

1. No investment is made for negligible games;

2. For the safe game predicted by the joint, buy the winning handicap of the strong team;

3. For risky matches, use Forward Operation (FO) to buy the winning handicap of the strong team, or Reverse Operation (RO) to buy the winning handicap of the weak team.

So based on the three forecasting methods, there are six investment methods, and the results of investment returns are shown in Table 2. Only subjective and objective joint predictions have the ability to screen out safe games from risky games, that is, games in which dark days and events will not occur, while subjective prediction and objective prediction alone will only know about risky games and negligible games. .

Table 2 Investment return results

It can be found from the table that the reverse operation investment strategy (SOP-RO) based on the joint forecast judgment index has the highest return.

According to the black swan event decision-making method based on subjective and objective joint prediction proposed in the feedback embodiment of this application, the first prediction data is obtained by obtaining predicted events and historical data sets; predicting the predicted events through the prediction model; in the preset subjective experience database Find the second predicted data that matches the predicted event; compare the first predicted data with the second predicted data for difference processing to obtain a predicted difference, and make a decision on the predicted event based on the predicted difference. Therefore, the reliability of predicted events is predicted through subjective and objective prediction to improve the performance of identifying black swan events and risk control.

Next, the black swan event decision device based on subjective and objective joint prediction proposed according to the feedback embodiment of the present application will be described with reference to the accompanying drawings.

As shown in FIG. 3, the black swan event decision-making device based on subjective and objective joint prediction includes: a first acquisition module 100, an objective decision module 200, a subjective decision module 300, and a decision output module 400.

The first obtaining module 100 is used to obtain predicted events and historical data sets.

The objective decision-making module 200 is used to predict the predicted event through a predictive model to obtain first predictive data; wherein the predictive model is generated by statistical analysis of historical data sets through a preset machine learning method.

The subjective decision-making module 300 is configured to search for second predicted data matching the predicted event in a preset subjective experience database.

The decision output module 400 is used to compare the first predicted data and the second predicted data for difference processing to obtain a predicted difference, and make a decision on the predicted event according to the predicted difference.

Further, in an embodiment of the feedback of the present application, making a decision on a predicted event based on the predicted difference includes:

Further, in an embodiment of the feedback of this application, the historical data set is historical data related to the predicted event within a preset period.

The second obtaining module is used to obtain multiple prediction data related to the prediction event;

The generation module is used to analyze multiple predicted data related to predicted events to generate a preset subjective experience database.

It should be noted that the foregoing explanation of the embodiment of the black swan event decision-making method based on subjective and objective joint prediction is also applicable to the device of this embodiment, and will not be repeated here.

According to the feedback embodiment of the present application, the black swan event decision-making device based on subjective and objective joint prediction is proposed by obtaining predicted events and historical data sets; predicting predicted events through a prediction model to obtain first predicted data; in a preset subjective experience database Find the second predicted data that matches the predicted event; compare the first predicted data and the second predicted data for difference processing to obtain a predicted difference, and make a decision on the predicted event based on the predicted difference. As a result, the reliability of predicted events can be predicted through subjective and objective prediction to improve the performance of identifying black swan events and risk control.

In addition, the terms "first" and "second" are only used for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first" and "second" may explicitly or implicitly include at least one of the features. In the description of feedback in this application, "multiple" means at least two, such as two, three, etc., unless otherwise specifically defined.

In the description of this specification, descriptions with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean specific features described in conjunction with the embodiment or example , The structure, materials or characteristics are included in at least one embodiment or example fed back in this application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine the different embodiments or examples and the characteristics of the different embodiments or examples described in this specification without contradicting each other.

Although the embodiments of the feedback of this application have been shown and described above, it can be understood that the above embodiments are exemplary and should not be construed as limiting the feedback of this application. Those of ordinary skill in the art are within the scope of feedback of this application. Changes, modifications, substitutions, and modifications can be made to the above-mentioned embodiments.

Claims

A black swan event decision-making method based on subjective and objective joint prediction is characterized in that it includes the following steps:

Obtain predicted events and historical data sets;

Predicting the predicted event through a predictive model to obtain first predictive data; wherein the predictive model is generated by performing statistical analysis on the historical data set through a preset machine learning method;

Searching for second predicted data matching the predicted event in a preset subjective experience database;

The first prediction data and the second prediction data are compared for difference processing to obtain a prediction difference, and a decision is made on the prediction event according to the prediction difference.
The method according to claim 1, wherein the making a decision on the predicted event according to the predicted difference comprises:

Determining that the predicted difference is less than a preset threshold, then determining that the predicted event is a security event;

If it is determined that the predicted difference is greater than or equal to the preset threshold, it is determined that the predicted event is a risk event.
The method according to claim 1, wherein the historical data set is historical data related to the predicted event within a preset period.
The method according to claim 1, wherein the historical data set is statistically analyzed through a logistic regression model of scikit-learn.
The method according to claim 1, further comprising:

Acquiring a plurality of prediction data related to the prediction event;

Analyzing the plurality of prediction data related to the prediction event to generate the preset subjective experience database.
A black swan event decision-making device based on subjective and objective joint prediction is characterized in that it includes:

The first acquisition module is used to acquire predicted events and historical data sets;

The objective decision-making module is used to predict the predicted event through a predictive model to obtain first predictive data; wherein the predictive model is generated by statistical analysis of the historical data set through a preset machine learning method;

The subjective decision-making module is used to search for second prediction data matching the predicted event in a preset subjective experience database;

The decision output module is configured to compare the first prediction data and the second prediction data for difference processing to obtain a prediction difference, and make a decision on the prediction event according to the prediction difference.
The device according to claim 6, wherein the making a decision on the predicted event according to the predicted difference comprises:

Determining that the predicted difference is less than a preset threshold, then determining that the predicted event is a security event;

If it is determined that the predicted difference is greater than or equal to the preset threshold, it is determined that the predicted event is a risk event.
The device according to claim 6, wherein the historical data set is historical data related to the predicted event in a preset period.
8. The device according to claim 6, wherein the historical data set is statistically analyzed through a logistic regression model of scikit-learn.
The device according to claim 6, further comprising:

The second obtaining module is used to obtain a plurality of prediction data related to the prediction event;

The generating module is configured to analyze the plurality of predicted data related to the predicted event to generate the preset subjective experience database.