Counting Cars in Parking Lots Became a Serious Investment Input

The Basic Idea

Companies report their results quarterly weeks after the end of the period. In between those reports investors make assumptions. Alternative data It is any information that narrows down that conjecture without waiting for the presentation and that does not come from the company itself

The categories are broad. Aerial and satellite images of parking lots marine terminals and construction sites. Aggregated and anonymized card transaction panels. Web traffic application downloads and job offers. Shipping manifests and customs records. Flow data by geolocation

What unites them is that neither was created for investors. A payment processor collects transaction records to move money a satellite operator photographs the terrain for agriculture and defense clients and a job board hosts postings so people can apply. The investment use is a secondary market in the escape of some other business which is the reason the offer exists and also the reason it is not reliable in the specific ways described below

Why It Works at All

Each of these is a proxy that is something related to the number you are interested in but not the number itself. Cars in a parking lot correlate with visits to the store which in turn correlate with sales. Job openings for warehouse staff correlate with expected volume

The value is in timeliness rather than accuracy. A rough estimate available six weeks in advance of the official figure can be worth much more than a precise figure that everyone receives simultaneously

The Chain of Assumptions

Each proxy depends on a chain of relationships that are maintained at the same time and the chain is longer than it seems

Take the parking lot example all the way. Cars in the parking lot must track visits which requires the number of people per car to be stable. Visits must track transactions which requires the conversion rate to be maintained. Transactions must track revenue which requires the average basket size to be stable. Revenue at the observed locations must track total revenue which requires that the sample of stores remain representative. And revenue must track the figure reported for the segment whichrequires that the accounting not be modified

Five links each of which is an assumption rather than a fact and any one of them broken is enough to reverse the signal. A company that increases prices opens a delivery channel closes stores in a region or changes its segment reporting has broken a link and none of those events are announced in the images

That's why professionals talk about understanding the collection methodology rather than the accuracy of the data. The data set can be perfectly accurate about what it measures and still be useless because what it measures and what is desired are linked by a chain of assumptions that were only tested once

The Problems Are Substantial

The difficulties are greater than the popular narrative suggests which is why most of these data sets disappoint

The coverage problem is the most common failure. A transaction panel that covers a small percentage of consumers concentrated on certain card issuers and revenue brackets can track a company's overall sales well for years and then diverge as the customer mix changes. Nothing announces that the relationship has broken down

The short history compounds this. Validating that a signal predicts outcomes requires many observations and a data set with three years of history offers approximately twelve quarterly data points which is not enough to distinguish a real relationship from a coincidence

The Composition Problem Is the Nastiest

Of the four the switch panel deserves separate treatment because it produces a false signal that is indistinguishable from a true one

A panel of card transactions is not a fixed group of people. Consumers are added and removed as the underlying data agreements change as issuers join or leave and as the provider expands into new regions. When the panel changes the measured spending changes with it and the change looks exactly like a change in consumer behavior

Worse the vendor has a commercial incentive to grow the panel so the trend is usually in the direction of greater coverage which shows up as apparent growth in what is being measured. A subscriber supporting testing with a panel that expanded over the period is a product of the vendor's sales success

The defenses are procedural rather than statistical. Insist on a dashboard that is held constant for historical comparisons request the data as it was seen at each point in time rather than currently recast and check to see if the vendor has ever reviewed the history. A data set whose past changes when you download it again cannot support backtesting at all

The Bar Is the Consensus, Not the Company

One point that reframes the entire exercise: accurately predicting a company's revenue is not enough and is not really the goal

The price already reflects an expectation. Analysts publish estimates those estimates are aggregated into a consensus and the stock moves based on the difference between the reported number and that expectation rather than the number itself. A data set that predicts earnings perfectly and predicts exactly what everyone already assumed produces no return

Therefore the amount being forecast is the surprise which is a much smaller and louder number than revenue. Getting within a percentage of a number is impressive as a measure and useless as a signal if the range of plausible outcomes the market is pricing in is narrower than that

This also explains why the same set of data is worth different amounts in different companies. It is most valuable when fundamentals are difficult for analysts to observe by other means when the reported figure moves the price and when coverage is sparse enough that the consensus is misinformed. In the case of a heavily covered company whose sales are already tracked by various industry sources the information arrives after its price has been set

The practical version of this is that a signal should be evaluated against the consensus estimate as it stood before the report not against the previous quarter and not just against the reported number. Evaluations that skip this step typically conclude that a data set works when what it has shown is that earnings are somewhat persistent

The Decay

The economics of this industry follow a predictable arc. The early adopters of a data set earn real excess returns. The data provider seeing demand sells to more customers because incremental sales cost almost nothing. As adoption spreads the information is incorporated into prices more quickly and the advantage shrinks to zero

What remains is not an advantage but a requirement. Once a data set is widely used not having it means being systematically late so companies pay for it to avoid a disadvantage rather than gain it. This is a significant and costly distinction

Why the Vendor Always Wins the Trade

Decay is not an unfortunate side effect. It is the predictable result of incentives on the part of sellers and it is worth seeing it clearly before signing anything

The marginal cost of selling a data set to one more customer is close to zero so each additional subscription is almost pure profit for the provider and a direct reduction in the value of the subscription for everyone who already has it. Exclusivity would preserve advantage and is costly precisely because the seller is asked to give up the rest of the market

What buyers usually get is limited exclusivity a limit on the number of subscribers or an early access period. Each of these slows deterioration rather than prevents it. The provider's incentive does not change and the contract is periodically renegotiated

The practical consequence is that a set of data should be evaluated based on how long its advantage is expected to last not how strong it appears today. A signal that works now and will be trading in two years is a two-year asset with a recurring cost which is a different investment case than the usual one

The Legal and Ethical Boundary

The limitation that shapes the entire industry is non-public material information.Trading on confidential information obtained in breach of duty is illegal and alternative data must clearly remain on the other side of that line

The general principle is that independently observed information about a company is legitimate while information that originates within the company or from its partners in breach of an obligation is not. Satellite images of a public parking lot are observational. A data set collected from a supplier's internal systems in violation of its contracts is a serious problem and the buyer assumes the risk

Consumer privacy adds a second limitation. Data derived from individuals requires adequate anonymization and consent and regulation here has been tightened considerably

The awkward feature of this limit is that the buyer often cannot verify which side a data set is located on. The chain from the original collection to the delivered file may pass through several intermediaries and the terms under which the data was first collected are not visible in the file. Therefore due diligence on the provenance of a data set and contractual representations on how it was obtained are an ongoing cost of operating in this area rather than a formality

The Cost Is Not the Subscription

The item is the license fee and is usually the smallest part of what it costs to use a data set

Each new source must be onboarded which means ingesting it mapping its identifiers to the company's own security master aligning its timestamps with everything else managing its revisions and creating monitoring so that a silent change in the source is noticed rather than absorbed. That's engineering work it's vendor-specific and it's repeated every time the vendor changes its format

Then there is the evaluation itself which consumes the researcher's time and in most cases ends in rejection. A company that evaluates many data sets a year will license a small fraction of them so the cost of those that were tested and discarded must be borne by the few that work

This is what makes alternative data a business of scale on both the buy and sell sides. The fixed cost of the evaluation process and apparatus is virtually the same whether you support a small fund or a large one and only the large one can spread it over enough capital to justify it

The consequence is worth stating clearly because it goes against how the field is typically described. The advantage of alternative data has gone from having access to unusual information to having the infrastructure to test a lot of it quickly and discard most of it. The data is for sale to anyone. The apparatus for figuring out which of them is worthless is not

What Good Practice Looks Like

Serious users treat these data sets as an input rather than a signal to trade directly. They test whether the relationship with reported results holds over many quarters and across many companies understand the collection methodology well enough to know when it would break and monitor for composition changes that silently invalidate a history

The failure mode is to adapt a story to a short story and trust it which is the same mistake as any overfit model arriving through a more expensive channel

The Bottom Line

Alternative data is a legitimate attempt to observe business activity directly rather than waiting to be told about it. The information is real the proxies are imperfect and occasionally fail without warning and the advantage erodes as adoption grows. It has largely become the cost of staying up to date rather than a way to get ahead. The questions that matter before you buy are how many assumptions lie between the measure and the number you are interested in whether the panel producing it has been stable and for how long the vendor intends tokeep the subscriber list short