Alternative Data is a revolution. More important than AI! p.1/2
The use of Alternative Data is the current super trend strongly shaping the present and future of investing.
It is a revolution more important than the revolution that Artificial Intelligence will give.
The main benefit of Alternative Data (AD) is that you can have most of your important metrics faster and more accurately than ever before.
Alternative Data currently improves signal quality and reduces risk. In the future, it will be the primary source of signals. Today, it is the main source of competitive advantage in many cases to find and use these signals before others.
In addition to the current use of AD, it is imperative to create competencies in the fund to use it wisely in the future because this area will now evolve rapidly.
Thanks to AD, the signal will be received earlier. As a result, it will be better, less risky, the position will be easier to run, and the exit will be better.
I have prepared twenty examples of using alternative data. I want to show a broad spectrum of situations where AD have been used so far and can be used in the future.
It is often the case that an exciting and inspiring idea comes from a completely unexpected direction. Therefore, it is worth attending industry conferences and collecting as many case studies as possible of what others have done.
Counting cars in the Tesla parking lot
One fund applied Machine Learning (ML) tools to analyze satellite images of the parking lot in front of Tesla's Megafactory. The tool analyzed the position of cars and their colours. The goal was to determine if and how the number of vehicles was changing. It turned out that cars quickly disappear and new ones appear in their place, which indicated that the company would keep its commitments and plans. Moreover, with this analysis, it was known weeks before the official announcement. Such information provides a substantial advantage over other market participants.
The above is a rather famous and awe-inspiring example, so it is worth commenting on it. If we look at the area around this factory (in Nevada), we can see one access road.
Setting up a car with a camera there and having someone count the vehicles leaving the parking lot would give the same result (and perhaps much cheaper) as using a whole team to machine-analyze satellite images. It is worth knowing this and looking for a way to have the same data not for a quarter of a million dollars but for 1% of that amount.
Non-Farm Payrolls Employment Data
A big fund noted the possibility of estimating NFP data with a high degree of precision by observing how many new listings from job seekers appear on a large website that connects workers with employers.
Using ML tools, they examined the relationship between the number of new job search postings and the subsequently published employment data. They found a formula that worked very well. It estimated the magnitude and direction of currency price movements a few days before the employment data was released. Rumour has it that traders found this way back in 2012.
Knowing in advance what the key data may look like allows you to build a position for the expected movement, maintain the existing position or exit it. Some traders (and many textbooks) recommend exiting the markets before the publication of the most critical data ("because the market may move hundreds of pips in any direction"). However, if we know what the data will be and the market's expectations are, then in practice, we have a money-making machine.
I recently read on Twitter the typical speculation about future NFP data and what the markets are expecting. But, unfortunately, there is already a group that knows and can exploit this.
Data on corporate flight usage valuable in predicting M&As
For companies investing in mergers and acquisitions, information about flights on jets leased by corporations is a good source of information about possible events. Two funds are cited as better-known examples. One of them made over 300 million, and the other 700 million using such data.
Traders tracked the flights of key people of a particular company to the city where another large company interested in an acquisition is headquartered. The board flight data was the first sign that something interesting might be going on.
When the information began to be confirmed, large positions were built. Prior knowledge allowed them to enter the market before everyone else.
Social media data can stand alone as a source of valuable signals.
Another big quant fund researched which analysts posting on Twitter had the best results and set up the following system. The application watches the Twitter feed of a selected group of top analysts and places orders when recommendations appear. The whole thing happens automatically.
Many analysts have spent years studying industries and sectors, and they have vast experience that they share on social media. Therefore, this strategy is a "no brainer".
Reading data from company documents with ML allows you to identify those companies that are likely to use creative accounting ("cooking books")
A trader uses this tool to pick out suspicious companies, creates a list of them, and further analyses the situation. As soon as his suspicions are confirmed, he builds shorts.
This example shows a system based on tracking anomalies in documents.
I have not heard of any fund doing this on a large scale, although they have similar tools at their disposal. They work in conjunction with other methods to give a better, more complete picture of a given situation.
Program enters a position by analyzing the news.
News moves the markets, but the sheer volume of information is a problem.
One fund created a stand-alone system to analyze news about a selected group of companies. The automat reads the news and analyzes the sentiment it contains. If it is positive, it buys; if negative - it sells.
Of course, there are additional elements here, but I want to show the very essence of the solution. Everything is done automatically. The trader is undoubtedly able to repeat some of these positions, but only a tiny part. The algorithm never sleeps. Hence it can work constantly and on every market from Tokyo to New York.
Moreover, with time, other factors will be added to the analysis of news sentiment. As a result, the strategy will be expanded and improved.
The capital will systematically flow to algorithms of this type. And this is not good news for discrete investors and traders.
The analysis of internet searches focused on sports brands shows the deepening weakness of the sector as well as the weakness of the biggest brands there.
Later, weakness has been confirmed by poorer performance, which in turn has brought price declines.
It's worth paying attention to this example - it shows trends in sentiment around companies and in the sector itself - in other words, changes in sentiment over time.
Soon, we, or rather the biggest traders and funds, will gauge sentiment trends for the major groups that consume a company's products and the sentiment of the major groups that own the company's stock. This insight will make the company picture clearer and the signals better.
The trader who has this valuable knowledge earlier will win.
Traders use new data while already in a position to extend their position or exit it.
A trader holding shares of a gaming company ordered a survey on whether the current customers would be willing to buy a new game that the company was developing. Most of them said yes, so he kept his position. The game turned out to be a success, and the price of the stake he held rose.
The survey was a way to forecast demand for the new product long before it was sold and long before the company's results were released.
This example shows using data not just as a signal but as information on whether it is worth extending a position.
Traders use AD to eliminate weaker signals.
One fund uses a mean reversion strategy. It is based on finding excess deviation from the average price of a company and opening the position hoping for a reversal.
For example, the price went too high, and the machine will try to catch the correction. Each element of the system, i.e. what is "too far", which moving average is the best, is determined with the help of statistics.
After publishing data on companies, this system enters the - it evaluates whether the upward movement is not too strong and trades on the declines.
If the amount of good data is high or the positive mood around the company lasts longer - the falls are weaker.
In this case, the program either does not enter the market or exits if it already has a position.
ADs allow you to understand better what makes up a signal.
One large fund analyzed the factors that influence the share price in the sector and found over 200 of them (they used PCA/ICA - Independent Component Analysis). They got 100 important ones and a sound system for predicting financial performance by removing the least significant.
The system examines incoming data and gives a forecast based on that, and there is no simple analytical relationship.
This is one example of a new type of system. "Manual" analysis of 100 factors would require a large team of skilled analysts and perhaps several years of work to find a working relationship. And then feeding the signal would maybe need a few additional days of work.
Today, building a system using Machine Learning is neither cheap nor short. However, in return, the finished system will analyze and give the result in seconds.
It is worth considering that the main factor that was analyzed was the quarterly results until recently. Today, 100 different factors can be analyzed simultaneously! First, it is worth examining which of the initially selected ones have a significant impact.
Does such knowledge give a market advantage? For example, suppose we ponder the capabilities of a trader or even a whole team of analysts and traders. Then juxtapose it with a system that automatically analyzes 100 factors every day for each company in a few seconds.
This news is not favourable for investors.
If you liked this post, give it a boost 🚀 and drop a comment so we know to publish more for you. Cheers!
Follow: www.tradingview.com/u/drSwierk/
It is a revolution more important than the revolution that Artificial Intelligence will give.
The main benefit of Alternative Data (AD) is that you can have most of your important metrics faster and more accurately than ever before.
Alternative Data currently improves signal quality and reduces risk. In the future, it will be the primary source of signals. Today, it is the main source of competitive advantage in many cases to find and use these signals before others.
In addition to the current use of AD, it is imperative to create competencies in the fund to use it wisely in the future because this area will now evolve rapidly.
Thanks to AD, the signal will be received earlier. As a result, it will be better, less risky, the position will be easier to run, and the exit will be better.
I have prepared twenty examples of using alternative data. I want to show a broad spectrum of situations where AD have been used so far and can be used in the future.
It is often the case that an exciting and inspiring idea comes from a completely unexpected direction. Therefore, it is worth attending industry conferences and collecting as many case studies as possible of what others have done.
Counting cars in the Tesla parking lot
One fund applied Machine Learning (ML) tools to analyze satellite images of the parking lot in front of Tesla's Megafactory. The tool analyzed the position of cars and their colours. The goal was to determine if and how the number of vehicles was changing. It turned out that cars quickly disappear and new ones appear in their place, which indicated that the company would keep its commitments and plans. Moreover, with this analysis, it was known weeks before the official announcement. Such information provides a substantial advantage over other market participants.
The above is a rather famous and awe-inspiring example, so it is worth commenting on it. If we look at the area around this factory (in Nevada), we can see one access road.
Setting up a car with a camera there and having someone count the vehicles leaving the parking lot would give the same result (and perhaps much cheaper) as using a whole team to machine-analyze satellite images. It is worth knowing this and looking for a way to have the same data not for a quarter of a million dollars but for 1% of that amount.
Non-Farm Payrolls Employment Data
A big fund noted the possibility of estimating NFP data with a high degree of precision by observing how many new listings from job seekers appear on a large website that connects workers with employers.
Using ML tools, they examined the relationship between the number of new job search postings and the subsequently published employment data. They found a formula that worked very well. It estimated the magnitude and direction of currency price movements a few days before the employment data was released. Rumour has it that traders found this way back in 2012.
Knowing in advance what the key data may look like allows you to build a position for the expected movement, maintain the existing position or exit it. Some traders (and many textbooks) recommend exiting the markets before the publication of the most critical data ("because the market may move hundreds of pips in any direction"). However, if we know what the data will be and the market's expectations are, then in practice, we have a money-making machine.
I recently read on Twitter the typical speculation about future NFP data and what the markets are expecting. But, unfortunately, there is already a group that knows and can exploit this.
Data on corporate flight usage valuable in predicting M&As
For companies investing in mergers and acquisitions, information about flights on jets leased by corporations is a good source of information about possible events. Two funds are cited as better-known examples. One of them made over 300 million, and the other 700 million using such data.
Traders tracked the flights of key people of a particular company to the city where another large company interested in an acquisition is headquartered. The board flight data was the first sign that something interesting might be going on.
When the information began to be confirmed, large positions were built. Prior knowledge allowed them to enter the market before everyone else.
Social media data can stand alone as a source of valuable signals.
Another big quant fund researched which analysts posting on Twitter had the best results and set up the following system. The application watches the Twitter feed of a selected group of top analysts and places orders when recommendations appear. The whole thing happens automatically.
Many analysts have spent years studying industries and sectors, and they have vast experience that they share on social media. Therefore, this strategy is a "no brainer".
Reading data from company documents with ML allows you to identify those companies that are likely to use creative accounting ("cooking books")
A trader uses this tool to pick out suspicious companies, creates a list of them, and further analyses the situation. As soon as his suspicions are confirmed, he builds shorts.
This example shows a system based on tracking anomalies in documents.
I have not heard of any fund doing this on a large scale, although they have similar tools at their disposal. They work in conjunction with other methods to give a better, more complete picture of a given situation.
Program enters a position by analyzing the news.
News moves the markets, but the sheer volume of information is a problem.
One fund created a stand-alone system to analyze news about a selected group of companies. The automat reads the news and analyzes the sentiment it contains. If it is positive, it buys; if negative - it sells.
Of course, there are additional elements here, but I want to show the very essence of the solution. Everything is done automatically. The trader is undoubtedly able to repeat some of these positions, but only a tiny part. The algorithm never sleeps. Hence it can work constantly and on every market from Tokyo to New York.
Moreover, with time, other factors will be added to the analysis of news sentiment. As a result, the strategy will be expanded and improved.
The capital will systematically flow to algorithms of this type. And this is not good news for discrete investors and traders.
The analysis of internet searches focused on sports brands shows the deepening weakness of the sector as well as the weakness of the biggest brands there.
Later, weakness has been confirmed by poorer performance, which in turn has brought price declines.
It's worth paying attention to this example - it shows trends in sentiment around companies and in the sector itself - in other words, changes in sentiment over time.
Soon, we, or rather the biggest traders and funds, will gauge sentiment trends for the major groups that consume a company's products and the sentiment of the major groups that own the company's stock. This insight will make the company picture clearer and the signals better.
The trader who has this valuable knowledge earlier will win.
Traders use new data while already in a position to extend their position or exit it.
A trader holding shares of a gaming company ordered a survey on whether the current customers would be willing to buy a new game that the company was developing. Most of them said yes, so he kept his position. The game turned out to be a success, and the price of the stake he held rose.
The survey was a way to forecast demand for the new product long before it was sold and long before the company's results were released.
This example shows using data not just as a signal but as information on whether it is worth extending a position.
Traders use AD to eliminate weaker signals.
One fund uses a mean reversion strategy. It is based on finding excess deviation from the average price of a company and opening the position hoping for a reversal.
For example, the price went too high, and the machine will try to catch the correction. Each element of the system, i.e. what is "too far", which moving average is the best, is determined with the help of statistics.
After publishing data on companies, this system enters the - it evaluates whether the upward movement is not too strong and trades on the declines.
If the amount of good data is high or the positive mood around the company lasts longer - the falls are weaker.
In this case, the program either does not enter the market or exits if it already has a position.
ADs allow you to understand better what makes up a signal.
One large fund analyzed the factors that influence the share price in the sector and found over 200 of them (they used PCA/ICA - Independent Component Analysis). They got 100 important ones and a sound system for predicting financial performance by removing the least significant.
The system examines incoming data and gives a forecast based on that, and there is no simple analytical relationship.
This is one example of a new type of system. "Manual" analysis of 100 factors would require a large team of skilled analysts and perhaps several years of work to find a working relationship. And then feeding the signal would maybe need a few additional days of work.
Today, building a system using Machine Learning is neither cheap nor short. However, in return, the finished system will analyze and give the result in seconds.
It is worth considering that the main factor that was analyzed was the quarterly results until recently. Today, 100 different factors can be analyzed simultaneously! First, it is worth examining which of the initially selected ones have a significant impact.
Does such knowledge give a market advantage? For example, suppose we ponder the capabilities of a trader or even a whole team of analysts and traders. Then juxtapose it with a system that automatically analyzes 100 factors every day for each company in a few seconds.
This news is not favourable for investors.
If you liked this post, give it a boost 🚀 and drop a comment so we know to publish more for you. Cheers!
Follow: www.tradingview.com/u/drSwierk/
