Of all the companies I’m betting against, my biggest bet is against Applied Digital (APLD).
I’m the portfolio manager of a small hedge fund, Fieldsong Investments, and we hedge our long equity positions with put options. These are essentially bets against stocks; they make money if the stock goes down in price more than a certain amount. Our hedge has more invested in Applied Digital puts than puts on any other company.
This article is devoted to discussing Applied Digital, its benefactor Nvidia (NVDA), and the detection of egregious manipulation of financial reports.
The History of Applied Digital
According to The Bear Cave, a newsletter devoted to exposing corporate misconduct, the company was incorporated in Nevada in 2001 as Reel Staff to provide staffing to production companies, but did a stock merger the next year with Flight Safety Technologies. Then in 2008 it pivoted to plasma technology and changed its name to Applied Science Products. Then it acquired Cummins Family Produce the next year so that it was now processing and packaging potatoes.
If that strikes you as crazy, this kind of switching is not that uncommon for shell companies. In 2021, the company changed its name to Applied Blockchain after acquiring crypto mining assets. The next year it changed its name to Applied Digital and switched to next-generation datacenters. All of this is frankly admitted on their website.
The company’s Chairman and CEO is Wes Cummins (of Cummins Family Produce). Cummins was also President of B. Riley Asset Management; B. Riley Financial, its parent company, recently made headlines due to the bankruptcy of Franchise Group, one of its key holdings, and is the most shorted stock in the financial sector. If you want to read more about both Applied Digital’s and B. Riley's shenanigans, here’s a good article.
But none of these are the reasons I bet against this company. I create ranking systems to guide me in choosing stocks to bet on (go long) and to bet against (using put options). Applied Digital was for many months ranked #1 on my ranking systems for puts, but I could never afford the options: I don’t like to pay too high a premium. They seemed like a stock certain to collapse, with terrible financials, negative price momentum, poor sentiment indicators, and barely any prospect of ever making any money. But then they raised $160M from a Nvidia private placement, which made the stock price go up and the put price go down. For me, this presented a good buying opportunity.
This is a company with negative gross margins, huge net losses, depleted cash reserves (well, until the Nvidia loan), large asset sales, equipment issues, and questions about the true demand for their supposed services. Last quarter’s interest expense was $18M, revenue was $44M, operating income was –$37M, net loss was –$65M. Those numbers were clearly unsustainable, and the most recent quarter’s numbers are significantly better, but their cost of goods remains higher than their revenue. In addition, they recently issued $53M in convertible preferreds, and have plans to issue another 49M more shares at $3.24 (the current price is over $6). There’s a good Seeking Alpha article that outlines much of this.
Why Nvidia Propped Up Applied Digital
Nvidia seems to be shoring up Applied Digital, infusing it with cash so that it doesn’t fail. They did this with CoreWeave last year. In a nutshell, Nvidia is loaning vast sums to its own customers so that they can buy more Nvidia chips. And those customers are drowning in debt, with hugely negative income, so they’re using those chips as collateral. How can this last? The scheme is unsustainable.
How is Nvidia increasing its revenue when the price of its chips is going down (indeed, all of Nvidia’s most popular chips have decreased in price) and why are those revenue increases so much higher than those of comparable companies like Advance Micro Devices and Taiwan Semiconductor? In part, it’s by making unsafe loans that effectively prop up its revenue. It’s basically selling chips to itself. Almost all of its sales increases are concentrated in sales to data centers and similar clients. Applied Digital and CoreWeave aren’t the only data companies with shaky financials: check out Super Micro Computer (SMCI), another Nvidia customer whose Beneish M-Scores (explained below) are super high. Once Nvidia’s customers start to default—and once they start buying non-Nvidia chips—will the ship go down? And how long will that take?
Earnings Manipulators (and Why Applied Digital and Nvidia Can Be Labeled as Such)
Last year I researched at length a system that estimates, from financial statements, the probability that a company is engaged in manipulation of their financials or “fraudulent” behavior, closely based on the work of Messod Beneish. Some of what I’m going to say here repeats what I wrote then.
At that time, an investment research firm exposed a company I had invested in (Tingo Group) as a massive fraud. Most of this company’s revenue apparently came from some other companies that they had recently acquired, but it turned out to be practically impossible to verify that any of that revenue actually existed outside of the company’s books. Shortly thereafter, the SEC suspended trading in its shares and filed charges against its founder.
Based on their most recent financial statements, I had given this company a very high rating and had put a lot of money behind it. (I wasn’t the only one to rate the company highly: Seeking Alpha offers “quant ratings” for all publicly traded US stocks, and gave it an A+.)
This hurt. Not only did it hurt financially—I lost over $80,000—but it hurt my credibility as a quantitative investor. In the big picture, I could stomach the loss. But my investing philosophy is founded on the maxim that safer stocks are better investments than risky ones. And yet I put a large bet on a stock that this research firm would shortly and rightly label a swindle with fictitious financials.
One reason I was caught off-guard is that Tingo Group was listed on the NASDAQ as a US company. Listed stocks domiciled in the US are the safest stocks on earth simply because the SEC is better at going after fraudsters in the US stock market than any other government regulatory body.
Stock market history, though, is full of frauds. It’s in the very nature of the beast. We are taught that we should beware of companies engaging in manipulation of their financial statements. Yet every company, by the very nature of financial reporting, has some leeway in constructing financial statements. There are hundreds of gray areas when reporting financials—which is one reason I frequently say that there is absolutely nothing scientific about financial data. What am I doing when I rank stocks based on this discretionary data, then? Well, I’m dealing in probabilities. If a stock’s reported free cash flow is high, that doesn’t mean that it’s a winner: it means that it’s more likely to be a winner than a stock whose reported free cash flow is extremely low.
So what could have tipped me off that Tingo was a fraud? Was there something quant-based that I should learn from this experience? Was there a screening rule I could apply to my universes to prevent this from happening again? Prompted by these questions, I researched and published the above-linked deep exploration into the subject of detecting fraudulent reporting.
In 1999, a professor of accounting at Indiana University published “The Detection of Earnings Manipulation” in Financial Analysts Journal. (This was a thorough revision of a method he had first published in 1997.) Messod Beneish was not the first nor the last to attempt to use financial statement analysis to detect fraud. But his method has held up better than any other; in a 2020 follow-up paper, “The Cost of Fraud Prediction Errors,” he proved that it had a lower ratio of false to true positives than any other method besides a machine-learning-based one introduced in 2020.
Beneish arrived at his formula by examining 74 firms that manipulated earnings over a six-year period and comparing those to 2,332 similar firms that did not. He came up with eight factors whose higher values seemed to indicate a greater chance of manipulation, seven of which compare the latest fiscal year’s figure to the previous one’s.
There are a lot of online versions of the Beneish M-score (M stands for manipulation), and many of them take slightly different approaches to the one Beneish originally proposed. The closest one available to Beneish’s original method is an excellent online calculator from Indiana University’s Kelley School of Business, where Beneish teaches, which breaks down any company’s M-score by factor and gives probabilities and odds. It’s faithful to the original calculation, using annual figures and Winsorizing, which most other services don’t do.
According to the Kelley School calculator, Applied Digital gets a very high score, with 120-to-1 odds that the company is a “manipulator” (meaning that there’s less than a 1% chance that it isn’t). Nvidia’s score isn’t quite so high, but it still gets 63-to-1 odds that it’s a “manipulator” too.
When I studied how the M-score works, I looked at the M-scores and the eight factors for about forty companies that have received damning reports from an investment research firm that specializes in accounting irregularities, undisclosed related-party transactions, and illegal financial reporting practices. Not all of those companies engaged in financial manipulation, but a majority of them seem to have.
As a result of my research, for which I used Compustat data on Portfolio123, I had the hubris to revise Beneish’s M-Score. Now, in addition to the Beneish M-Score, I can look at a Revised M-Score for any company. Both Applied Digital and Nvidia get extremely high revised scores too: Applied Digital is in the top 100 out of the Russell 3000 and Nvidia is in the top 150. (If you’re a subscriber to Portfolio123, you can check both the original and revised score of any stock here.)
Digging Into Applied Digital’s and Nvidia’s Financial Statements
The M-Score is made up of eight different measures, seven of which compare the most recent fiscal year’s numbers to those of the previous year.
Applied Digital gets the highest possible score on the first of these measures: days sales in receivables. This is the ratio of receivables to sales, which can be converted to the percentage of sales for which cash has not been received. If that ratio is getting bigger, it’s a sign that the company may be overstating its sales. Nvidia’s score on this measure is also somewhat high, but not egregiously.
The second measure looks at the company’s gross margin. If it’s deteriorating markedly, that gives it a strong incentive to manipulate its financials. But what I found is that if a company’s gross margins are wildly increasing, that’s also plenty of cause for concern. Applied Digital gets a very low score on Beneish’s original measure, but a very high score on my revised M-Score, because it has been increasing its gross margin by a truly improbable amount. (Nvidia’s gross margin has also been increasing, but not to such an extent.)
The third measure concerns the balance between a company’s current assets, net plant, and total assets: it looks at the change in the ratio of non-current assets that are not a part of net PPE to the company’s total assets. If that number is going way up, the company may be improperly deferring costs—or engaging in substantial acquisitions. Applied Digital, once again, has very high numbers on this measure. Nvidia does not.
Both companies, though, score extremely high on the next measure, which is simply sales growth. Beneish found that high-growth firms are more likely to engage in financial statement fraud; in my experience, this is the number-one flag for revenue falsification. And indeed, as I pointed out earlier, Nvidia is recording revenue increases by selling chips to companies and then using those chips as collateral to lend those companies sufficient money to buy their chips. It’s quite clear that at least some of Nvidia’s revenue increases are due to the way they are more or less buying chips from themselves—and earning interest from doing so.
Nvidia scores very high on the fifth measure as well, which is the change in the ratio of depreciation to the sum of depreciation and net plant. If this ratio is going down quite sharply, that means that the company may have been revising upwards their estimation of the useful lives of their assets, thus increasing the company’s net income through financial manipulation.
Both companies have very high scores on the ratio of SG&A to sales. A huge increase in sales without a corresponding increase in SG&A is a clear warning sign that sales may be overstated, and that’s abundantly true for both of these companies.
The seventh measure is an odd one: the change in leverage, as measured by the ratio of total debt to total assets. While Beneish reasoned that if debt is increasing, there would be more incentive for earnings manipulation, his numbers found the opposite, and he ended up assigning this factor a negative coefficient. In this case, Applied Digital is increasing its debt quite dramatically, while Nvidia is decreasing its debt quite dramatically (in part by increasing Applied Digital’s debt). My research confirms Beneish’s numbers: a dramatic decrease in the ratio of debt to total assets can often signify that a company is increasing their equity by selling lots of shares. One of the most prevalent signs of a company engaging in fraudulent activities is that they sell shares.
Beneish’s last measure is the ratio of total accruals to total assets. Once again, Nvidia scores very high on this measure, which makes sense considering how it’s accounting for its chip sales.
In my revised M-Score I added one more measure, which examines the increase in fully diluted shares, because I noticed that a lot of fraudulent companies issue a lot of new shares or issue additional treasury shares or sell convertible preferreds to make money before they’re caught. And they aren’t always the same companies as those with low scores on the seventh measure (debt to assets), since many of them both issue shares and increase their debt load. Applied Digital, which does both, gets a very high score here.
Conclusion
There is no simple method for catching financial fraud, just as there is no simple method for picking winning stocks. However, paying attention to certain signs of fraud would have helped me avoid the Tingo Group fiasco. And I have my fingers crossed that Applied Digital will soon suffer the same fate as the Tingo Group: its share price will go down to almost zero.
As for Nvidia, I have few such hopes. Is it engaging in earnings manipulation? It seems like it to me. Is it a house of cards, like Enron, bound to collapse because of it? I doubt it. But I do expect it to be a much, much smaller company, in terms of its market cap, in a year or two, once investors realize how it has inflated its reported revenues.
My CAGR since 1/1/2016: 41%.
My top ten holdings right now: PSIX, ELMD, KINS, ARREF, ESP, BKTI, DSP, FLXS, LUGDF, HMENF.
