In this article, we will construct the prosperity index of the sub-sectors of the transportation industry, and at the same time, we will construct a timing strategy based on the valuation performance of the industry, and try to use the transportation prosperity to capture the industry's earnings in the future.
summary
Industry index timing: Starting from the degree of industry prosperity, the nonlinear impact of industry prosperity on index returns is depicted
Use the constructed prosperity index to select the timing of the industry index. There are many factors that affect the performance of industry index returns, including fundamentals, valuation changes, market conditions and many other factors. Among them, the degree of prosperity depicts the fundamental state of the industry, and the timing of the industry led by the prosperity can obtain stable fundamental support, reduce the risk of long-term volatility and drawdown of the strategy, and provide investors with relatively stable returns. In this article, we will construct the prosperity index of the sub-sectors of the transportation industry, and at the same time, we will construct a timing strategy based on the valuation performance of the industry, and try to use the transportation prosperity to capture the industry's earnings in the future.
The economic logic and hypothesis testing of the industry are used to screen the industry prosperity indicators. Based on the business logic of each transportation industry, we establish two tests for the indicators that conform to the economic logic: 1) whether the indicators can explain the growth rate of the industry's net profit, and 2) whether the indicators explain the net profit and match the changes in index returns. For the indicators that pass the test, we further screen the indicators based on the data coverage and logical importance, and synthesize them into the industry prosperity index.
Innovation point 1: Distinguish different market stages and highlight the contribution of the boom. We combine the classification tree and the Probit model to distinguish between different market stages, and predict and analyze the returns of industry indices. Compared with the general forecasting framework, this model has two advantages: 1) by distinguishing the market stage, the model can better adapt to the nonlinear role of indicators in the industry index forecast, and 2) with the help of the model to identify the contribution of industry prosperity and valuation indicators in the model forecast, which can screen out the market signals with high economic contribution and fundamental support, and improve the timing effect of the model.
Innovation point 2: Use price percentiles as the model prediction target. We use the quantile substitution yield of the current price of the industry index over the next 40 days as the forecast target of the model. Compared with the yield index, quantile prediction has the following advantages: 1) quantile can make full use of the price information of each day in the future, reflecting the overall trend and volatility of the index in the future period; 2) quantile prediction has less sensitivity to the holding period, which is more in line with the market's lagged response to changes in industry prosperity and has a certain degree of stability.
Taking the port industry as an example, the test set has a long-time return of 9.8% and a winning rate of 87.5%
The prosperity of the port industry is mainly affected by throughput. The port industry business revolves around shipping, the unit price and cost of services are relatively stable, and the net profit is mainly affected by the shipping throughput. In the construction of the index, we group 41 industry indicators according to four types of economic logic: throughput, shipping demand, shipping price and congestion, and finally select the three indicators of "coastal total: total of major ports in China: container throughput year-on-year: current month value", "year-on-year growth index of China's import value" and "CTFI: composite index" to synthesize the industry prosperity index.
The excess return of the long timing model test set reached 7.1%. We took 2017.1.1 – 2021.12.31 as the training set and 2022.1.1 as the test set, and used the above-mentioned prosperity index and industry valuation indicators to test the port industry index at a timed basis. Compared with the industry index, the annualized volatility and maximum drawdown of the long-short timing result decreased from 20.1% and 17.8% to 13.4% and 10.5%, respectively, the return increased from 2.7% to 9.8%, and the excess return reached 7.1%. In terms of long-short timing, the model sends 10 signals outside the sample, with an annualized return of 8.3% and a volatility of 16.6%.
The prosperity timing model has stable cross-industry returns. At the same time, we conducted an out-of-sample timing test on the post-2022 data of five transportation sub-sectors, including shipping, road, railway, aviation and airports. The five industries were 3.6%, 5.6%, 2.7%, 16.4% and 17.2% respectively.
Risk Warning: The quantitative models constructed in this report are based on historical data statistics, and there may be a risk of failure when the industry characteristics or market environment changes in the future.
body
Industry Prosperity: An indicator of medium- and long-term earnings
Industry Prosperity: Characterize the profit trend and assist the income expectation
What is the industry prosperity? Industry prosperity is the portrayal of the fundamental state of the industry and the future change trend. In the field of investment, investors expect to buy stocks in related industries during the upward stage of the industry to obtain more lucrative returns, and sell related stocks during the downward stage to avoid possible risks. This report attempts to construct an industry prosperity index through relatively high-frequency data, which can help us achieve real-time tracking of the fundamental status of the industry and provide fundamental-level reference for investors to make investment decisions.
Research Objectives of Industry Prosperity: As we analyzed in the report "Fundamental Quantitative Series (5): How to Quantitatively Track Cyclical Industry Prosperity", the rise and fall of industry indices is the result of many factors, including fundamental factors, transaction-level factors, financial market liquidity and other factors, and the industry prosperity research focuses on the fundamental-level factors. Therefore, we believe that the construction of the industry prosperity model should be based on the profit forecast, and the ultimate goal is to realize the prediction of the return of the industry index.
► Profit forecast: When the economy is up, the overall profit growth rate of the industry should be improved, and conversely, when the prosperity is weak, the overall profit growth rate of the industry should also decline (or even show negative growth). The use of prosperity indicators to achieve profit forecasts is a validation of the fundamental logical framework, which can help investors find economic variables that have a greater impact on the prosperity of the industry.
► Industry index income forecast: The use of prosperity indicators to achieve the prediction of industry index returns is the application scenario of industry prosperity research, investors can take the research results of industry prosperity as an important reference basis for medium and long-term allocation, especially in the case of relatively stable financial market liquidity and no major changes in investor sentiment, the influence of industry prosperity changes may be enhanced.
Chart 1: Factors influencing sector index returns
Source: CICC Research
Figure 2: Objectives of the Industry Prosperity Study
Source: CICC Research
How to build industry prosperity: take the port industry as an example
How to filter valid economic indicators?
The explanatory strength of the industry's net profit growth rate is the basis for constructing the industry prosperity index. This chapter mainly discusses how to start from the industry indicators and screen out the relevant indicators that can effectively reflect the overall prosperity of the industry. Industry indicators are essentially descriptions of the fundamental characteristics of the industry, and adjust the performance expectations of the industry by reflecting changes in demand, supply, price, etc., thereby providing fundamental support for the medium and long-term performance of the industry. Therefore, effective industry indicators should have an explanatory effect on the medium- and long-term earnings performance of the industry index, and this explanatory effect is achieved by changing the industry's net profit growth expectations. In the following article, we will construct three steps for the validity screening of a single indicator based on the economic indicators of the platform, and screen the effective indicators of the corresponding industry:
► Data preprocessing: readjust the missing values of some indicators in the industry indicator pool according to the publication date, standardize the transformation, and mark the economic logic of the role of the indicators.
► Explanatory test: Use the Granger test to determine whether the economic logic represented by each industry indicator is explanatory to the growth rate of industry net profit, and if it passes the test, it is considered that the indicator is an effective indicator to measure the fundamentals of the industry.
► Isotropic test: test the correlation direction between the indicator and the change in income and net profit of the industry index, if it is the same, it is considered that the indicator can provide effective information through the basic industry timing.
Figure 3: Single-metric effectiveness screening process
Source: CICC Research
Through frequency adjustment and standardization, different metrics are given a unified metric. This section mainly discusses the pre-processing process of industry indicators, including confirming the economic logic of indicators, adjusting missing values according to the publication date, and standardizing the indicators.
Update the data date to remove the impact of individual data storage time. Due to the late storage time of some indicators, the publication date of historical data in the database is the storage date, which is inconsistent with the actual publication time. In this regard, we use a threshold of 35 days (the publication date of the main data is generally one month after the data statistical date, considering the impact of non-working days, taking 35 days can avoid quoting future data), if there is a data in the data that contains a statistical day span of more than 6 months, it is considered that the data has a storage problem, and the actual publication date is adjusted to 35 days after the deadline of the statistical date and the data publication date is the former.
Normalize all values to -1~1. The original values of the indicators are standardized to facilitate the synthesis and interpretation of the index later. For year-on-year indicators, we take the original values of the indicators and truncate the extreme values above 100% and below -100%. For other indicators, the quantile of the current indicator in the past 5 years is taken and monotonically narrowed to -1 to 1. Therefore, the value range of all indicators is -1 to 1, and a value between 0 and 1 indicates that the fundamental information represented by the indicator may be upward, otherwise, the corresponding indicator is in a downward phase.
Figure 4: Data preprocessing process
Source: CICC Research
Subjective estimation of the direction of the effect of indicators on the economy. The changes in the industry economy are cyclical, and the direct use of statistical tests to test the influence direction of the indicators may cause spurious regression problems due to phase differences. According to the economic logic of the index, we give a priori judgment on the direction of the industry prosperity of the index, which provides a subjective basis for the follow-up test.
Monthly frequency tests the ability of indicators to explain the growth rate of net profit. In order to avoid the differences caused by the different frequency of industry indicators and facilitate the comparison between indicators, we uniformly adjust the frequency of indicators and net profit growth to the month, and explain the changes in industry net profit from April to December in the next 4 to 12 months with the indicator value at the end of the month.
The Granger test is used to determine the explanatory effect of the index on the growth rate of industry net profit. The Granger test is a method of analyzing the causal relationship between time series of economic variables, in which X is considered to be able to explain Y if the exogenous variable X still has an explanatory effect on the current value of Y when the endogenous variable Y is included in the past. In addition, the Granger test requires the sequence to be stationary, so we perform a unit root test on the indicator before testing. In this report, we take 10% as the threshold, and if the significance of the test coefficient is lower than this level, the indicator is considered to have passed the explanatory test.
Judge whether the interpretation of the indicator on the growth rate of net profit is consistent with expectations. As mentioned above, changes in the industry's prosperity are cyclical, and the interpretation of the indicators for the prosperity will also be affected by the phase. If the explanatory effect of the indicator on the growth rate of net profit leading from April to December is significant in the expected direction, it is considered that the indicator has a strong ability to explain net profit.
► Estimating the direction of impact: According to the economic logic of the index, subjectively judge the direction of its impact on the growth rate of the industry's net profit.
► Periodic adjustment: Adjust the year-on-year/month-on-month growth rate of each industry indicator and industry net profit TTM to a monthly basis, and use the indicator value of the latest announcement date for data with a frequency higher than the monthly one.
► Causality test: The industry indicator is used to perform the Granger causal test on the growth rate of the industry's net profit from April to December in the future, and whether the explanation of the indicator on the year-on-year/month-on-month net profit TTM is significant (<10%).
► Confirmation of the interpretation direction: Determine whether the interpretation direction of the significant indicators in the causality test is consistent with the expected impact direction.
Indicators that are explanatory to the growth rate of net profit are not necessarily price guided. In the explanatory test, we use the Granger test and the economic logic check to screen out the index groups that have a robust explanatory ability for future industry economic changes. However, there are differences in the predictive power of different indicators on the future earnings of the industry.
Use the correlation of indicators to price changes to filter valid indicators. We test the price correlation of the indicators that have passed the explanatory test, and require that the correlation between the index and the industry index is the same as the interpretation direction of the index on the industry's net profit growth, and further screen out the effective indicators that can fully reflect the fundamental information to the changes in the industry index. Considering the volatility of the sector index itself, we also tested the correlation between the indicator and the changes in the sector index and the price quantile over a period of time in the future to provide more robust results for the test. In the next section, we take the port industry as an example to show how to start from the raw data of the industry and filter out a combination of single indicators that can reflect the changes in the industry's prosperity.
Screening effective indicators of the port industry: taking into account the industry logic and single-indicator test performance
The prosperity of the port industry is mainly affected by the throughput logical chain. Overall, the port business revolves around ocean freight, service prices are relatively stable, and revenue growth is largely dependent on throughput. In the long run, the correlation between the net profit of the industry and foreign trade is strong, and the increase in trade throughput will enhance the net profit realization of the port in the case of the overall upward trend of the world economic environment. Therefore, we estimate that the prosperity of the port industry is also positively correlated with the value of imports and exports and the macroeconomic environment measured by the PMI of major markets.
Figure 5: Framework for analyzing the prosperity of the port industry
Source: CICC Research
The port industry indicators are divided into four types of economic logic: throughput, shipping demand, shipping price and congestion. We selected 41 indicators of the port industry, and according to the economic analysis framework shown in the figure above, we started from the throughput logical chain and divided the 41 industry indicators into four types of economic logic: throughput, shipping demand, shipping price and congestion.
Explanatory test: 26 valid indicators were selected from 41 industry indicators. We performed an interpretive test on 41 indicators in the port sector, which removed samples from the test set (2022 to date) to avoid introducing future information. When the p-value of the expected impact direction of the indicator in the year-on-year or month-on-month test of net profit is less than 10%, we believe that the indicator has an explanatory effect on net profit, and finally we screen out 26 effective indicators from 41 industry indicators.
Isotropy test: 25 indicators passed the test. As mentioned in the previous section, in addition to analyzing the indicator's explanatory power to the fundamentals of the industry, it is also necessary to consider the indicator's ability to predict the return of the index. In this regard, we took 40 trading days as the expected holding period to test the impact of the publication of the corresponding indicator values on the return of the industry index. In order to avoid the impact of short-term price fluctuations, we calculated the absolute return of the industry index in the 40 days after the release of the indicator data and the percentile of the index in the next 40 days on the announcement date. After testing, we screened out 25 single indicators.
Construction of port industry prosperity index: combined with industry logic, reflecting the timeliness of indicators
In the previous section, we introduced the general test framework of the industry prosperity index and selected a set of valid single indicators using the port industry as an example. In this section, we will select a single indicator that we believe is more effective from each underlying economic logic and carry out the final synthesis of the prosperity index.
We select their representative indicators from the three robust logics of demand, throughput and freight rate: "China: import value: year-on-year in the current month", "total coastal value: year-on-year in the total container throughput of major ports in China: current month" and "CTFI: composite index", and synthesize them into the port prosperity index. Considering the test performance and logical representativeness of the indicators, we used three indicators to synthesize the port industry prosperity index: "China: import value: year-on-year growth in the current month", "total coastal value: year-on-year increase in container throughput of major ports in China: current month" and "CTFI: composite index". The synthesis method is that after the three indicators are standardized, they are filled in according to the publication date, and the average of the three equal weights is taken every day as the industry prosperity index. Among them, since the explanatory effect of "CTFI: Composite Index" on the net profit of the industry is negative, its weight is also adjusted accordingly.
Chart 6: Port Industry Sentiment Index
Note: Data as of April 8, 2024
Source: Wind, General Administration of Customs, China Port Network, CICC Research
The application of industry prosperity index in the timing model
By combining industry prosperity index and market indicators, the industry prosperity model provides timing signals for industry allocation. Industry fundamentals is an aspect of analyzing industry returns, and the return performance of industry indices is the result of the combined effect of fundamental factors and market factors. In the previous chapter, we mainly started from the basic indicators of the industry, carried out inspection, screening and logical combing, found out the indicators that can effectively explain the fundamentals of the industry, and constructed the industry prosperity index. For market factors, we will use the P/B and P/E valuations of the industry to reflect the impact of trading sentiment and macro liquidity.
Based on the industry prosperity index and valuation data, we establish an industry timing model based on classification tree and Probit model. By inputting the daily industry prosperity index, industry valuation data and industry index prices, the model can predict the price trend in the next 40 trading days (the percentile where the current price is in the 40 trading days), and send corresponding timing signals when there is economic information to support it.
Figure 7: Timing process of industry prosperity model
Source: CICC Research
Industry timing model: Combining the classification tree and the Probit model, the industry prosperity and valuation status are attributed
The classification tree and Probit model were used to provide explanations for the prediction results. As mentioned above, the role of the prosperity index in different market stages is different, and this difference can be better identified by using classification trees. In addition, the Probit model can provide a linear explanation for the role of business indicators within the same market stage.
The classification tree is used for market stage identification, and the Probit model is used for intra-stage prediction. Considering the heterogeneity of the industry prosperity index in different market stages, as well as the characteristics of the two types of models themselves, we combine the classification tree and the Probit model. We first use the classification tree to identify the market stage of the industry, and then apply the Probit model internally at the different stages of the identification to predict the percentile of the current price. In order to avoid the problem of overfitting, we take the single indicator of "year-on-year growth rate of China's monthly import value" as the industry prosperity index, and use the classification tree, Probit model and the joint model of the two to change the long-short timing of the port industry index CI005358.WI. The chart below shows the relative net value of the three (benchmarked against the Port Industry Index CI005358.WI), which is an overall improvement over the use of the classification tree and the Probit model alone.
Figure 8: Comparison of model timing results
Note: The long-short timing results of each model in the test set are shown here, and the net value is 1 on the start date of the test set on January 1, 2022, and the data is as of April 8, 2024
Source: Wind, General Administration of Customs, China Port Network, CICC Research
Characterize future price fluctuations. The yield only takes into account the price of the current and future point in time, and does not take full advantage of the price information in the intermediate period. The percentile indicator can make full use of the price information of each day in the future, reflecting the overall trend and volatility of the index in the future period.
When the effect period of industry prosperity on index returns is uncertain, the adaptability of price quantile indicators is stronger. The market's understanding of the fundamentals of the industry is based on the overall trend of economic changes. In different market stages, the intensity and lag time of the prosperity index will change, and the fixed holding cycle rate of return is easily affected by the fluctuation of the boom cycle. In this report, we use the quantile where the index price will be in the next 2 months as the forecast target. Compared with the yield indicator, the quantile uses all the information of the next 40 trading days, and does not limit the specific holding period, which has better adaptability in the case of periodic fluctuations in the role of the indicator.
Percentile indicator prediction reduces the risk of drawdown in the short term and has better out-of-sample adaptability. The following table shows the timing results of the port industry using the yield index and percentile index, with "China's monthly import value year-on-year growth rate" as the industry prosperity index. Compared with the yield indicator, the percentile indicator has a significant improvement in the return and maximum drawdown in the test set.
Figure 9: Comparison of timing results of different indicators
Note: The relative net value here = the long-short net value of the port industry index of the corresponding index/port industry index, the relative net value and the port industry index are both taken as the net value of the test set (2022.1.1), and the data is as of 2024-4-8
Source: Wind, General Administration of Customs, China Port Network, CICC Research
Train a classification tree model of industry prosperity. The first part of the model starts from the market stage prediction and divides it into 3 layers, 7 classification tree root/branch nodes and 8 leaf nodes. Each of the seven root/branch nodes of the classification tree corresponds to a segmentation index and a segmentation threshold, so that the error of predicting target y in the left and right sub-nodes after splitting according to the threshold is minimal. The 8 leaf nodes correspond to 8 sets of regression models based on market stages, including the correlation coefficients of prosperity and valuation indicators, and the corresponding constant terms of the stage itself.
Identify nonlinear relationships with multiple leaf nodes. In terms of parameter selection, we require that the number of leaf nodes should exceed the number of potential market stages, so as to improve the ability to identify nonlinear effects. As mentioned above, the boom may have a nonlinear effect on the industry index, in addition to the difference in direction in different market stages, there may also be marginal effect changes within the same stage, and fitting with an appropriate amount of leaf nodes can better adapt to this characteristic.
Figure 10: Classification tree model framework and parameters
Source: CICC Research
Figure 11: Optimal conditions for classification tree model training
Source: CICC Research
Identify the signals that are dominant in economic change. As mentioned above, the forecast results are affected by many factors, such as the market stage, prosperity and valuation of the industry. Compared with the short-term effect of valuation changes, opportunities led by economic changes are supported by fundamentals, and earnings are more stable. Therefore, when sending out signals, in addition to requiring the forecast target to show a better timing opportunity, it is also required that the prosperity contribution plays a more prominent role in the contribution to the change of the forecast value.
Attribute model predictions to market stage, boom, and valuation. We divide the change in the daily forecast value relative to the previous trading day into three parts: 1) period contribution: the same prosperity and valuation state, resulting in differences in the model's price percentile forecast in different market stages, 2) prosperity contribution: the change in the prosperity index in the same market stage, and 2) valuation contribution: the forecast change caused by valuation in the same market stage. The above three items are expressed as 1) data of the previous period * (parameters of the current period - parameters of the previous period), 2) (prosperity of the current period - prosperity of the previous period) * parameters of the current period, 3) (valuation of the current period - valuation of the previous period) * valuation parameters of the current period, and the sum of the three exactly constitutes the change of the daily forecast value relative to the previous trading day.
According to the price quantile forecast and the change of prosperity, the timing signal is issued. We use 0.5 as the validity threshold to signal bullishness when the price percentile forecast is greater than 0.5 (the estimated percentile of the current price in the next 40 days is within 25%) and the prosperity contribution is greater than the highest value in the last 5 days. Conversely, if the price percentile forecast is below -0.5 (the estimated percentile of the current price in the next 40 days is higher than 75%), and the economic contribution is high, a bearish signal is issued. At the same time, new signals are not repeated for dates that have already been signaled within 40 days.
Timing model construction of port industry: annualized rate of return to 9.8%
The change in the economy under the appropriate state of valuation is the main timing signal of the port industry. The following diagram illustrates the visualization of the predictive model making decisions. The vertical axis from -1 to 1 represents the forecast of the model with the corresponding cross-section input (e.g., -1 indicates that the current price is at 100% in the next 40 days and should give a bearish signal). From the structural point of view, the model's summary of the timing law of the port industry can be divided into three categories: 1) when the valuation is at a low level (the valuation indicators represented by EP and BP are at a high level), the industry index is easy to hover at a low level; 2) when the valuation is at a high level, the industry index is affected by the valuation and prosperity indicators The volatility is large, but the overall trend rises with the rise of the economy; 3) when the industry valuation is moderate, with the upward trend of the economy, the industry index has a stable upward path.
Figure 12: Visualization of forecasting model decisions in the port industry
Source: Wind, General Administration of Customs, China Port Network, CICC Research
The long-short and long-long timing net returns of the port industry timing model reached 8.3% and 9.8%, respectively. The port industry timing model takes the equal weighted mean of "China: import value: year-on-year growth of the current month", "coastal total: total container throughput of major ports in China: current month" and "CTFI: composite index" as the industry prosperity index, and the mean of the 5-year quantile of the industry index EP and BP as the valuation index, and uses the data from January 1, 2017 to December 31, 2021 as the training set for training, and conducts an off-sample timing test for the industry index from January 1, 2017 to the present.
In the timing operation, the buy/short selling operation is mainly to send a bullish/bearish signal (the forecast value >0.5, <-0.5, and the points that meet the requirements of the economic contribution will also trigger buying/short selling). In the case of an open position, the position will be closed when the percentile forecast breaks through 0 (0 for long positions, or 0 for the nonversa). The test results are shown below.
Exhibit 13: Net value of the full sample of the port sector
Note: Data as of April 8, 2024
Source: Wind, General Administration of Customs, China Port Network, CICC Research
Exhibit 14: Timing performance statistics of the port industry
Note: Data as of April 8, 2024
Source: Wind, General Administration of Customs, China Port Network, CICC Research
Article source:
This article is excerpted from: "Fundamental Quantitative Series (16): How to Quantitatively Characterize the Prosperity of the Transportation Industry ?——Based on the Characteristic Index Database of CICC Dianjing", which was released on April 12, 2024
古翔分析员 SAC 执证编号:S0080521010010 SFC CE Ref:BRE496
Yujie Cao Analyst SAC License No.: S0080524020013 SFC CE Ref: BUA485
Chen Yiyun Contact SAC License No.: S0080122080368 SFC CE Ref: BTZ190
Xiaoxiao Zhou Analyst SAC License No.: S0080521010006 SFC CE Ref: BRA090
Junwei Liu Analyst SAC License No.: S0080520120002 SFC CE Ref: BQR365
Zhang Xiaohan Contact SAC License Number: S0080122070155
Liu Gangxian Analyst SAC License No.: S0080520080003 SFC CE Ref: BOK824
Qikun Wu, CFA Analyst SAC License No.: S0080521020002 SFC CE Ref: BQI397
Wang Yixuan Contact SAC License Number: S0080122070135
Feng Qibin Analyst SAC License No.: S0080521090003 SFC CE Ref: BRW011
Legal Notices