CHAPTER 8
Social Media and the Internet of Things
- —What is the funniest joke a trading program can make?
- —Hello world.
The news that we see, read, and hear was traditionally chosen by editors based on its newsworthiness. Press releases allowed companies to share news that might not make it into a story. Today, the Internet takes over, with everyone, even inanimate objects, capable of broadcasting on the World Wide Web. Sponsored content sits next to news, and blogs by reporters and the public update continuously. The resulting content creates data series that, when processed in real time, can be used to derive valuable insights.
An event like the Brexit vote helps to illustrate the importance of social media and social media analytics to risk management. On June 24, 2016, the United Kingdom voted to withdraw from the European Union by a margin of 52 percent voting to “leave” to 48 percent voting to “remain.” While polling leading up to the event was nearly even, polls had moved toward the “remain” camp in the final weeks. Such was the perceived sentiment regarding the outcome that the leaders of the “leave” camp were essentially conceding defeat even before the polls closed. As the votes were counted, the “leave” vote began to emerge, and as the evening wore on and all of the polling stations reported their results, “leave” proved to be the outcome. The following morning, the prime minister resigned and a period of political and economic uncertainty began.
Investment banks had ordered staff to be on notice for such an outcome and traded through the evening. The British pound fell to its lowest point since 1985, the year before financial deregulation saw the city of London emerge to be the financial capital of Europe. Instantly, it was unclear what would be the future for financial firms in a country that would be separate from Europe. What securities could be bought and sold in London? Would foreign staffers be welcome?
With the world watching this event, the instant change in the valuation of currencies and stocks is not surprising. The financial markets quickly assessed this vote as bad news. Additionally, the actions of politicians to quit the scene in the following days added to the uncertainty and to the turmoil in the markets.
What is interesting, however, is how some social media analytics firms were predicting a “leave” vote even in the face of largely “remain” coverage screaming from the media outlets. What was the secret of the prediction? The answer is the frequency of “leave” versus “remain” mentions. Apparently, “leave” news stories, however negative with respect to their subject, appeared much more frequently than “remain” news. Several research studies indicated that the popularity of a particular news item, not the underlying sentiment, is the strongest predictor of future value of the item.
Similarly, research shows that stocks that are mentioned most often on social media are not only likely to experience volatility, but also are about to rise. Why does this make sense? At a certain level, the “all publicity is good publicity” rule holds. According to the classical marketing theory, for a consumer or investor to buy into an asset, several things must take place in the following order: awareness, consideration, purchase, retention, and advocacy. Awareness develops independently of whether the attached sentiment is positive or negative. In fact, marketing research shows that many people, when building awareness, store a neutral sentiment in their brains. In the consideration stage, the prospects are evaluating whether or not to purchase a product or a stock. At this stage, products that have higher awareness feel familiar, and are more desirable, independent of whether the sentiment that generates this awareness was good or bad. People appear to erase all judgment and only recall whether they have heard about a particular stock or idea before, and not whether what they heard was positive or negative.
In higher‐level terms, however, the core answer to the question of why news distribution matters is “psychology,” a field that never had a chance in finance until very recently. Until as little as 10 years ago, economists completely ignored the distribution of information and people's perceptions about it, including feelings, emotions, and reactions outside of the cold‐hearted positive versus negative number‐driven universe. While some research existed in the space of behavioral economics, it was largely considered to be a fringe activity. Available data‐processing speeds also mattered: In the more traditional approach where people are making trades based on daily‐data, intraday news was usually too short‐term to matter. Finally, little data existed to study the actual diffusion of news and opinions—social media was just not as advanced as it is now, and many academic experiments involving people's perception and opinion formation in response to news simply lacked data.
In other disciplines, like electrical engineering and psychology itself, however, information diffusion has been studied for generations. Electrical engineering succeeded in quantifying the actual transmission properties of news and any other information in real and near‐real time. Psychology has specialized in human processing of data, in‐group versus out‐of‐group opinion formation and the like. With the data afforded by social media, finance academics are now joining their electrical engineering and psychology colleagues in droves to assess the ultimate holy grail of financial economics: exactly how does news make some market participants place their money on the line to buy or sell?
The psychology of finance now extends well beyond pure trading. Some recent studies show that regulators, political commentators, and even top politicians themselves are considerably influenced by social media, and not always for the better. Some researchers insist that psychology is even the root cause behind financial crises: people under stress are thought to make the markets irrational and, in extreme cases, nonexistent. (Ironically, the same researchers stop short of endorsing robots on the trading floor.)
The important aspect of today's social media “news” is its descriptive properties. Search engines and other databases archive not just the news itself but the source, the context, the reaction of commenters, the impact on the news readers (indexing such as “most viewed” or “trending” are examples), and the proliferation of news through the social network of primary readers—are all meticulously time stamped and organized.
AbleMarkets Social Media Index, for example, confirms these findings, as does Ravenpack in pure news. By counting news items ahead of Brexit, and finding that “leave” news items way outstripped “remain” stories, Ravenpack claims to have predicted the Brexit vote. Tracking the real‐time mentions of specific financial instruments and keywords across a wide swath of social media, AbleMarkets Social Media Index is reliably predictive of short‐term returns and volatility.
AbleMarkets Social Media Index tracks Internet discussions concerning individual financial instruments continuously, and is consistently leading changes in prices. Figure 8.1, for example, is showing the daily AbleMarkets Social Media Index for AAPL and the corresponding daily closing prices for AAPL. As the figure shows, APPL closing prices follow or lag AbleMarkets Social Media Index.
Figure 8.1 AAPL in social media leads AAPL closing prices.
The impact of social media is, of course, palpable intraday in addition to the following day.
What properties are common to price changes that can be anticipated in social media? One property that is particularly pronounced across years of data signals is the stronger effect of social media news on smaller stocks. As multiple researchers point out, smaller stock prices are more likely to display social media reactions. After all, most investors are busy tracking the latest largest cap and SPY developments, leaving smaller stocks to fend for themselves. Thus, a “smaller stock” can be any stock outside of the S&P 100 range, not just a penny stock. As an illustration, Figure 8.2 presents AbleMarkets Social Media Index for VMware, Inc. (NYSE: VMW), and the increased predictability of price peaks is immediately apparent.
Figure 8.2 Normalized social media conversations, as measured by AbleMarkets Social Media Quotient (left axis) vs. same‐day intraday range volatility for VMware (ticker VMW)
Some research goes further and weighs the social media spectrum by the subject matter credibility of the commentator. Computers score social media users, web sources, and blog publishers on the selected topic in a superficial performance attribution analysis. They ask questions like “how well did each commenter predict past developments?” This methodology allows researchers to identify the more likely sources of data. Thus, for instance, AbleMarkets has deselected Twitter from its source pool altogether in 2013 as it became a noisy source with little predictive value as a whole.
Can Twitter users be segmented into the better‐informed and the chattering class? The tricky part of Twitter is its openness to programmatic interfaces. In other words, it is reasonably easy to create a software robot that will pluck content from the Internet and post it on Twitter. Why would one choose to do that without close supervision? Anecdotal evidence suggests that Twitter's human followers appreciate feeds with more voluminous and timely content, and they spend little time analyzing the quality and relevance of more distant posts. As a result, when analyzing Twitter, it is the machines that tend to fall into a trap. If you assume that all content is created by the same person (and not another robot randomly posting cursory‐related items on one's twitter feed), the machines end up assigning arbitrary scores to Twitter participants.
SOCIAL MEDIA AND NEWS
The first iPhone was launched on June 29, 2007, and the world hasn't been the same since. The speed and convenience with which we now process information and communicate has created a new sense of urgency to understand and participate in further unbridled conversations, collectively known as social media.
Still, even social‐media‐savvy traders and investors cannot read and analyze hundreds of news items that may be simultaneously produced for a multitude of financial instruments across the markets. A category of sentiment analytics sprung up to help investors deal with the onslaught of information. Essentially, computers count the number of positive and negative words occurring in a news article or a social media post. In posts related to financial markets, positive words include “happy” words like increase, expansion, and profit. Negative words comprise terms like losses, declining, and recession, among many others. Companies like Ravenpack scan through entire dictionaries, classifying words into “positive,” “negative,” and “neutral” categories. The total sentiment is calculated as a score, awarding points to positive words and subtracting points whenever negative words are encountered. When the number of positive words in an article exceeds the number of negative words, the article is deemed to have a positive sentiment. When the opposite holds, the sentiment is considered to be negative. In addition to figuring out whether a given news story contains good or bad news, Ravenpack finds a place for the story in its vast classification database. Does the article contain references to “AAPL,” the stock‐market ticker for shares of Apple, Inc.? If yes, the Apple bin gets an added entry. Does the article talk about “macroeconomics”? If so, the macroeconomics column is updated to reflect the news addition. SENSENews is another service that collects and semantically filters fundamental information about companies. SENSENews also looks at sentiment around the data and isolates 16 different investment criteria around the stocks it monitors.
Social media further affects the way people and machines read news. For example, machine processing of a streaming Dow Jones feed is interesting but may not be enough to qualify the reporter's understanding of the matter. That's where social media comes in. Scanning the website where the same Dow Jones article was published, and particularly zeroing in on the comments attached by often‐anonymous users following the article, one may deduce a so‐called agreement score. An agreement score is an independent crowd‐sourced data point that indicates to what extent readers really value the content of the piece. Often, the comments turn out more insightful than the article itself—and the agreement score sheds light onto the reality of the article subject.
In the Internet‐enabled world, all readers can express how they feel. Doing so is possible with abbreviations (LOL = laugh out loud) and “emoticons”—the cutesy little yellow faces often far from smiling, just to name a couple. However, many of these sources of human feelings may be misleading. LOL alone may refer to opposite emotions: happiness and sarcastic hostility. How do computers distinguish how people feel when they post?
Internet Sentiment
How Fintech Decides How You Are Feeling with Natural‐Language Processing
The innovation causing computers to read text generates reams of unstructured data. And with the data, a question for researchers: if you could choose, which of the following options would you prefer?
- Perform a large‐scale analysis that involves reams and reams of data.
- Type in a single question into a specialized search engine and let the engine do the analysis for you.
If you are like most people spoiled by the modern conveniences of Google, you would probably prefer option #2. The ability to understand a natural‐language question and convert it into machine‐readable instructions for an analysis is a field onto itself, and is known as natural language processing. This new discipline is at the heart of businesses like Kensho, backed by Google and Goldman Sachs.
Natural language processing (NLP) is not just limited to queries. Indeed, most of the advances of NLP are applied in the space of processing news articles, social media posts, and the like.
How does NLP work? The methodology can resemble sentiment analytics, but with its own quirks. Instead of maintaining tables separating positive words from negative words, NLP may instead maintain tables discriminating between positive and negative phrases. A phrase can be defined as a sequence of two or more words.
In addition, NLP researchers often assign probabilities that specific phrases will occur in natural language. When a phrase is used frequently than normal in a given piece of text, this alone may serve as a positive or negative indication of sentiment. NLP may similarly assign probabilities to the words following a specific phrase, and deduce sentiment based on what actually transpires. The study of NLP is complex and well developed in the field of Computer Science.
How does NLP work in practice? Most scientists begin the language‐parsing process by removing punctuation and words like the. Twitter‐like hashtags, searchable keywords preceded by “#” that are now popular on other social media like LinkedIn and Facebook, are converted to normal words with the hashtag removed. Next, all remaining characters are moved to lowercase for consistency, and so‐called terms—technical items in natural language processing—are born.
Processing terms is no easy task. First, to reduce computer confusion, the words are “cleaned up”—most often reduced to the 7,315 words contained in the complete modified Harvard IV‐4 dictionary. The remainder of text is often discarded. Some biases may result from this pruning operation alone. For instance, some financial slang, such as BOT and SLD which are old‐school trading acronyms for bought and sold, do not make it through the Harvard filter, among many others.
Next, some researchers choose to perform stemming—a procedure whereby all remaining words are reduced to their stem. Stemming may or may not work, depending on the end application. Separating positive and negative words often removes prefixes necessary for understanding the sentiment like “de‐,” “un‐,” “pro‐,” and so on. For example, stemming delivers equal‐probability positive/negative estimates for words like communication (normally, positive) and ex‐communicated (negative sentiment).
The words, taken in sequence, are next subjected to Latent Semantic Analysis (LSA)—a principal component analysis of sentences that determines the most relevant terms in the data pool. The idea behind the technique is that the text about a particular subject contains a large number of synonyms of the subject itself. The output “needles in the hay” form the basis for classification of the given opinion piece. In addition, clustering techniques can subsequently be used to find related articles, messages, or tweets to map a comprehensive media sentiment across various media channels.
Finally, the machines arrive at the final classification of sentiment, and here the algos follow one of two paths preferred by their human masters: “supervised” or “unsupervised” learning. Supervised learning really refers to learning by the book—the machines are provided dictionaries of positive and negative words, against which the outputs of LSA and clustering are screened. The unsupervised learning lets machines run wild in their own universe, delivering clusters of key data outputs. While the process may seem elaborate, it may take as little as a few microseconds from raw data to finished output, given proper machine configuration.
Are machines making bulletproof decisions on the news? Not always. A classic counterexample to artificial intelligence is a 2008 case of the following nature. An individual looks up an article in the Orlando Sentinel in the middle of the night, at 1:36 AM on Sunday, September 7, 2008. The article, written six years prior in 2002, detailed that United Airlines may file for bankruptcy. Surprisingly, the old article made the “most viewed” list at the Orlando Sentinel, and was picked up by Google as a trending story. Next, a Bloomberg reporter in search of content looks up the Google trend index, ignores the article's publication date, and instead writes a current article about the imminent bankruptcy of United Airlines. Markets open, and the United Airlines stock plunges 76 percent (!) before recovering by the end of the trading day, when the news was disentangled and straightened out. Of course, in this story, it was really a person, not a machine, making the final mistake. However, the computer‐generated risks of similar nature exist.
Where does social media evolve from here? The technology is already here to track not just what you type, but how you type it as well. The current frontier in sentiment analytics is parsing what you type in your reader comments. Basic, very fast, and accessible technologies like HTML5 already allow website designers to track and record how fast you type (a potential indicator of the strength of your emotions), how and where you move your computer mouse on screen, where you click, and so on. While you may take the information you read onscreen for granted, there is a price to pay in giving up your personal feelings and related data, and the price will only become more expensive day by day, as more and more data will be gathered and analyzed by the websites you visit.
Government Data
Social media helps make macroeconomic figures that much more accessible. Why should a researcher wait for the summary of the latest trends observed by the government compiled monthly by a team of government staffers? Why can't financial researchers themselves access the granular data as it becomes available to the government?
New York City is one of the first cities in the world to open much of the data it archives to the public. Available at https://nycopendata.socrata.com, the data contain over 1,600 time series, including noise complaints, taxi service, parking tickets, and more. What good is this information to someone whose job is to deduce the state of the economy? Take parking tickets, for example. While an individual ticket might not signify much, the aggregate changes in parking tickets may serve as an indicator of the economic health of New York City. More parking tickets may reflect the increases in parking shortages, which, in turn, is indicative of more demand for parking spaces, more consumer demand, and, ultimately, higher consumer confidence levels. Naturally, creativity, data analysis skills, and understanding of economics are required to translate seemingly obscure data into actionable economics signals.
Take trash, for instance. It has long been known that one man's trash is another man's treasure, but in the age of big data, the statement takes on a new meaning. It can be shown that NYC trash, and Manhattan's trash in particular, can be indicative of the impending economic changes. Thus, an increase in Manhattan's trash reliably predicts a decline in the S&P 500 as far as three months later. The predictive power may be due to corporate downsizing: trash generated by liquidating offices may only translate into the S&P 500 prices three months later. Whatever the relationship, it is statistically significant and can be used to generate profitability. At least one multibillion‐dollar hedge fund has a group devoted to analyzing and processing NYC data to generate incremental returns!
Timeliness of Data Analysis
The key to successfully using any sort of data is timely analysis. Take the NYC trash data, for example. Who reaps the rewards from the information? Usually, it is the person who obtains the data first and is the most capable of deriving conclusions from it.
With real‐time tasks come real‐time risks. Government data, for instance, are notoriously prone to revisions and updates. Social media data may be subject to Internet outages, malicious hacks, and other activity that puts a dent into objective data handling. Even the Internet of things data may become corrupted: what if the chip‐reading sensors fail?
How does one manage risks like that? The answers may boil down to:
- Redundancies
- Increased data sampling pools
- Averaging
Redundancy refers to multiple copies, often of literally everything from data sources, databases, power, to processors. Redundancy helps deal with temporary outages by transferring the load temporarily to an alternate solution, while the primary (failing) solution is awaiting recovery. If and when a key node in the data acquisition and processing chain should fail, the system enters a “fail‐over” mode whereby the rest of the units engage and take over the load. The fail‐over essentially defines what has become known as a cloud with all the complexity abstracted to the user.
Increasing the sources of data helps with redundancy. Suppose you harvest all of your data from the NYC website and the NYC mayor decides to unilaterally shut the data website down, tearing your business to shreds. Planning for situations like this involves robust number of parallel data sources that can be used together to determine a given outcome. Then, should one of the data sources disappear, the business will not stop, but will continue with other data inputs.
To further minimize the risk of disappearing data sources, multiple related data inputs can be treated as independent samples and ultimately combined into an average. Techniques vary from using rudimentary arithmetic averages to more sophisticated weighted versions. The resulting average will sustain the production of your analysis and diffuse the impact of the missing source should one of the underlying sources disappear.
Some readers may roll their eyes at the mention of averages. However, averaging and the underlying sampling are gaining increased attention in today's world where the amount of available data is outrageous. Take Twitter studies, for instance. Many researchers writing about Twitter's significance don't analyze every single tweet, as Twitter messages are called. Instead, they select 10 to 20 percent of the message universe, sampled in a way that is representative of the Twitter's true much larger universe. Done correctly, this technique helps derive meaningful inferences from mountains of data in a fast and productive manner.
THE INTERNET OF THINGS
Some 30 years ago, the Internet barely existed. People communicated using telephones and whirring fax machines, and the latter were so common that Manhattan ran out of 212 area code numbers. Since then, the Internet became not just widely available, but also a necessity for many. Forget millennials—you can see modern two‐year‐olds demanding to use iPhones from their parents!
As a result, the Internet has changed the way people do things, including how they trade, receive, and process news, and even game the system. This section discusses the resulting fintech disruptions.
Some social media is tailored to finance. Investors share their perspectives on certain chat sites that are dedicated to them. Chat rooms on Bloomberg, Twitter, or on websites created a new source of information and perspective on the reactions of traders to news. Social media evolved as well to having sites that polled investor sentiment and do so more and more frequently. Twitter created an ongoing stream of posts that could label and categorize names of companies.
Still, the world is filled with meaningful data that falls outside of traditional investment analysis. For instance, pricing information on comparable products is a big factor in detailed fundamental analysis, but such data was typically not readily available until recently. With the expansion of fintech, companies like ThinkNum deliver competitive intelligence by visiting thousands of retail websites at once, compiling product and pricing information. The information is then aggregated and converted into pricing and supply and demand benchmarks, and is used by investment professionals to gauge relative advantage of particular firms.
The process of accessing the websites electronically and recording the content of what humans normally see even has its own technical term: scraping. Of course, the scraped web interfaces are not completely oblivious to the advantage that scrapers receive from the posted data. The web interface providers, in turn, take measures to restrict scraping by identifying machine or repeat behavior. This cat‐and‐mouse game is complex, and successful scraping requires oodles of effort to deploy.
Tracking Shipments, Fleets, and Supplies in Real Time
The intelligence relevant to investment analysis is not limited to entries posted on the Internet. Many of the key figures, like supply and demand numbers traditionally assessed through polls and surveys, can now be reliably aggregated from the merchandise itself.
Take, for example, universal product codes (UPCs) we are completely accustomed to—bar codes attached to virtually everything we buy, and scan at the cash register for easy processing. By now, many items shipped or stored also contain their own radio‐frequency identifiers (RFIDs). An RFID is a “smart” UPC. Each RFID is a tiny electronic device that, when in the range of a radio‐frequency reader, reflects to the reader its unique identifier information. Most RFIDs are passive devices, in that they do not require their own energy source. Instead, the devices are activated by the energy transmitted by the RFID readers. Much of this technology was invented for the purposes of being undetectable in spying devices during the Cold War. Since they typically don't contain a power source, RFIDs emit no heat until activated by the RFID reader, and are, therefore, undetectable by conventional device scanners under most circumstances. Without a power source makes these devices long‐lived, potentially lasting indefinitely, and costless to maintain—there is no need to replace batteries or provide air conditioning. Aside from accidental breakage, the modern RFIDs are weather‐ and tampering‐proof, and are manufactured on a large scale.
RFIDs are not a rare species. Every time you buy new clothes or walk through a modern US pharmacy, you encounter RFIDs embedded in boxes of most items as deterrents to theft (the devices are activated by the scanners at the exit to the store and set off the alarms, unless previously deactivated at the cash register). While the devices are commonplace, it's the emerging ability to collect and to use this information that is innovative.
The best part of RFIDs is that the information they contain, as well as their geospatial coordinates, can be continuously scanned and stored. This information can be subsequently mined for various applications: Amazon uses RFIDs to optimize its warehouse shelf stocking, farms track livestock movements, and fleets follow the locations of their containers.
For financial professionals, all of these databases translate into information‐driven profits. The data on warehouse movements can be distilled into relative demand figures for IBM products versus those of Intel, for example. The quantities and movement of livestock allow for more accurate commodity futures pricing. Shipping container movements not only predict the financial health of shipping companies but also of fuel demand and a host of auxiliary products and services.
And the RFIDs are only bound to proliferate further, creating vast pools of information on moving cargo, cars, and other objects.
In addition, RFIDs and similar sensors allow us to build smart devices that continuously report to us their condition. For instance, a bridge built with smart cement (cement with embedded chips) can inform us of important cracks and other structural problems, before any major problems occur. For financial analysis, this translates into a new universe of previously unattainable data helping determine valuation of cement companies and the like.
The Internet of things (IoT) is only getting started. According to industry analysis, only 8 percent of businesses are using more than a quarter of the IoT data they generate. The opportunity for revenue growth from IoT is tremendous—consumers expect their smartphones to do more in an IoT‐enabled world and are ready to pay for it; development tools for IoT are continuously streamlined and simplified, and regulation is coming of age to prevent IoT abuses.
Internet of Things Adoption
What IoT devices are coming online in the near future? How about 300 million utility meters, 150 million cars, 1 million vineyard acres, to start?1 The upcoming widespread adoption of IoT across all consumer segments is anticipated due the following five trends in the markets:
- Smartphone expectations
- Businesses using more data
- Regulatory facilitation of IoT
- Advanced network connectivity
- Better network security
Many people today live and die by their phone—buzzing text messages, video calls, and a slew of other features. Phone companies compete with one another on the latest gleaming app waiting to induce millions of consumers to ditch their existing device in search of the newest, shiniest item. As such, phone users expect their phones to deliver increasingly more of everything: Turn on the heat in the house, please, and don't forget to water the plants when the soil is dry—are just a few of the basic examples of what is now possible to do remotely over the phone. The benefits generated by devices such as home automation are tremendous, and industries like advertising can use them to create not just targeted ads but targeted lifestyle solutions. Financial companies may, in turn, use the data to better understand consumer trends, predict supply and demand, and forecast the prospects of industries.
Similarly, businesses are prompted to examine the data they generate and take better care of the information. Many businesses are now studying how they can use their own data. From reorganizing shelves to selecting appropriate floor surfaces to tailoring lighting in their employee environments, businesses are able to incrementally improve performance and drive profitability using IoT. Also, IoT helps businesses better understand the usage of their products and build ever more sophisticated offerings.
Tracking user information with IoT, of course, is something that needs to be strictly regulated. Where do we draw a line on privacy in an environment where IoT information is emanating from countless user devices? As regulation of IoT evolves, it is bound to clarify and facilitate all aspects of IoT.
Network connectivity is projected to evolve to support more data, faster, more reliably, and at a lower cost and energy requirements. Already, the buzz is all about 5G—the fifth generation of cell phone technology that promises to deliver interconnected self‐driving cars, robots for multiple uses, as well as virtual reality on your phone.
Finally, security is continuously evolving to deliver protection to billions of device users across the globe. The more secure the devices, the more trust businesses and consumers will place into operating them, and the more useful the data output will be.
IoT is on the rise, and it is here to stay. While the beginnings of IoT were largely supported by early adopters, by now IoT is a recognized force across industries and enterprise scales. And business‐oriented IoT is apparently growing much faster than IoT for consumers—after all, the gain to business productivity can be large‐scale and simply phenomenal, while the people's adoption of IoT may take a while and be much more incremental in scope and value. Of course, financial companies have noticed and are already processing IoT data in search of an investment edge.
How does someone acquire the Internet of things data? There are numerous ways to do so. First, new technology allows for the direct identification of every single shipment and every piece of merchandise. Thanks to bar codes, every mailed item can be tracked online. The data about the origin of an item and the customer who buys it can be also sold to third‐party data distributors, enabling end users to analyze and derive inferences about various aspects of the economy way ahead of formal monthly and quarterly indicators. Also, consumer‐tracking data are backed directly by consumer orders and dollars, and not just by consumer opinions given to cold‐calling survey operators. As a result, observations of consumer purchases (what consumers actually do) are much more credible than what they say.
Other firms collect and sell data displayed on the Internet. In our age of electronic commerce, pretty much every firm has a website that displays its mode of operation, product descriptions, and, often, prices. This information can be collected and used for competitive analysis, and in finance, for fundamental analysis and cash projections for each entity. Complex computer programs are built to “scrape” and distill the information in a fast and efficient manner.
Such analysis is hardly new: In the past, consulting companies chartered planes to fly over factories and count the number of shipping containers as an indicator of supply and demand generated by a particular manufacturer. Today, such analysis can be performed at a fraction of a cost by analyzing satellite imagery with image‐recognition software, often involving neural networks.
CONCLUSIONS
Social media and the Internet of things disrupt the traditional model of how information is disseminated. The news about supply and demand of individual businesses or commodities, for instance, is no longer confined to quarterly reports, expensive consulting surveys, and newspaper articles. Instead, the news is generated continuously with smart devices and people. Computers process the news in real time or near real time and incorporate it into the markets. Whoever does not possess this new class of information risks being left behind.
END OF CHAPTER QUESTIONS
- How is social media changing the financial landscape?
- What is a news sentiment? How is it determined?
- What is the Internet of things?
- Which industries produce Internet of things content?
- What risks do investors face with social media and the Internet of things?