AI's Winners, Losers and Wannabes: An NVIDIA Valuation, with the AI Boost!

I will start this post with a couple of confessions. The first is that my portfolio has held up well this year, in a market that has been top-heavy and tech-driven, and one big reason is that it contains both NVIDIA and Microsoft, two companies that have benefited from the AI story. The second is that much as I would like to claim credit for foresight and forward thinking, AI was not even a speck in my imagination when I bought these stocks (Microsoft in 2014 and NVIDIA in 2018). I just happened to be in the right place at the right time, a reminder again that being lucky often beats being smart, at least in markets. That said, NVIDIA’s soaring stock price has left me facing that question of whether to cash out, or let my money ride, and thus requires an assessment of how the promise of AI play’s out in its value. Along the way, I will take a look at the promise of AI, as well as the perils for investors, drawing on lessons from the past.

The Semiconductor Business

    The semiconductor business, in its current form, had its growth spurt as a consequence of the PC revolution of the 1980s, as personal computers transitioned from tools and playthings for geeks to everyday work instruments for the rest of us. In the last four decades, computer chips have become part of almost everything we use, from appliances to automobiles, and the companies that manufacture these chips have seen their fortunes rise, and sometimes be put at risk, as technology shifts.

1. From High Growth to Maturity!

    It was the personal computer business in the 1980s that gave the semiconductor business, as we know it, its boost, and as technology has increasingly entered every aspect of life, the semiconductor business has grown. To map the growth, I started by looking at the aggregated revenues of all global semiconductor companies in the chart below from 1987 to 2023 (through the first quarter):

Source: Semiconductor Industry Association

From close to nothing at the start of the 1980s, revenues at semiconductor companies surged in the 1980s and 1990s, first boosted by the PC business and then by the dot-com boom. From 2001 to 2020, revenue growth at semiconductor businesses has dropped to single digits, as higher demand for chips in new uses has been offset by loss of pricing power, and declining chip prices. While revenue growth has picked up again in the last three years, the business has matured.

2. Sustained Profitability, with Cycles!

    The semiconductor business has generally been a profitable one for much of its existence, as can be seen in the  aggregate margins of companies in the business below:

While gross and operating margins have always been healthy, the pick up in both metrics since 2010 is a testimonial to the higher profitability in some segments of the chip business, even as competition commoditized other segments. As can be seen in the periodic dips in profitability across time, there are cycles of profitability that have continued, even as the business has matured. 

    It is worth noting that these margins are understated, because of the accounting treatment of R&D as an operating expense, instead of as a capital expenditure. The R&D adjusted operating margin at semiconductor companies is higher by about 2-4%, in every time period, with the adjustment to operating taking the form of adding back the R&D expense from the year and subtracting out the amortization of R&D expenses over the prior five years (using straight line amortization).

3. Love-Hate Relationship with Markets!

    As the semiconductor business has acquired heft, in terms of revenues and profitability, investors have priced those operating results into the market capitalization assigned to these companies. In the graph below, I report the collective enterprise value and market capitalization of global semiconductor companies, stated in US dollar terms:

As you can see, the semiconductor companies have enjoyed long periods of glory, interspersed with periods of pain in markets, starting with a decade of surging market capitalizations in the 1990s, followed by a decade in the wilderness, with stagnant market capitalization, between 2000 and 2010, before another decade of growth, with market capitalizations surged six-fold between 2011 and 2020. Note that for the most part, semiconductor companies carry light debt loads, leading to enterprise values that either trail in market capitalization in some years (because cash exceeds debt) or are very close to market capitalization in other years (because net debt is close to zero). 

    As market capitalizations have risen and fallen, the multiple of revenues that semiconductor companies has also fluctuated, reaching a high in the dot-come era, with semiconductor companies trading collectively at more than seven times revenues to a long stretch where they traded at between two and three times revenues, before spiking again between 2019 and 2021. If prices are a reflection of what the market thinks about the future, the pricing of semiconductor companies seems to indicate an acceptance on the part of investors that the business has matured.

4. Shifting Cast of Winners and Losers!

    As the semiconductor business has matured, it has also changed in terms of both the biggest players in the business, as well as the largest customers for its products . In the table below, we show the evolution of the top ten semiconductor companies, in terms of revenues, from 1990 through 2023, at ten-year intervals:

The cast of players has changed over time, with only two companies from the 1990 list (Intel and Texas Instruments) making it to the 2023 list. Over the decades, the Japanese companies on the list have slipped down or disappeared, to be replaced by Korean and Taiwanese firms, with Taiwan Semiconductors being the biggest mover, moving to the top of the list in 2022. After a long stretch at the top, Intel has dropped back down the list and ranked third, in terms of revenues, in 2022. Note that NVIDIA, the subject of this post, was eighth on the list in 2023, and has remained at that ranking from 2010. That may seem at odds with its rising market capitalization but it is indicative of the company's strategy of going after niche markets with high profitability, rather than trying to grow for the sake of growth.

    The customers for semiconductor chips have also changed over time, with the shift away from personal computers to smartphones, with demand emerging from automobile, crypto and gaming companies in the last decade. Over the last few years, data processing has also emerged as demand driver, and it is safe the say that more and more of the global economy is driven by computer chips:

Semiconductor Industry Association

The forecasts for the future (2030), were for faster growth in automobile and industry electronics, but the potential surge in demand from AI products was largely underplayed, showing how quickly market forecasts can be subsumed by changes on the ground.

NVIDIA: The Opportunist!

    NVIDIA was founded in 1993 by Jensen Huang, but it remained a niche player until the early parts of this century. Much of its rise has come in the last decade, just as revenues for the overall semiconductor business were starting to level off, and in this section, we will look through the company's history, looking for clues to its success and current standing.

1. Opportunistic Growth, with Profitability

    NVIDIA went public in January 22, 1999, with the dot-com boom well under way, and its stock price popped by 64% on the offering date. At the time of its public offering, the company was money-making, but with small revenues of $160 million, making it a bit player in the business. As you can see in the graph below, those revenues grew between 2000 and 2005, to reach $2.4 billion in 2005. In the following decade (2006-2015), the annual revenue growth rate dropped back to 7-8% a year, but that growth allowed the company to make the top ten list of semiconductor companies by 2010. Well-timed bets on gaming and crypto created a surge in the revenue growth rate to 27.19% between 2016-2020, and that growth has continued into the last two years:

There are two impressive components to NVIDIA's history. The first is that it has been able to maintain impressive growth, even as the industry saw a slowing of revenue growth (3.97% between 2011-2020). The second is that this high revenue growth has been accompanied not just with profits, but with above-average profitability, as NVIDIA's gross and operating margins have run ahead of industry averages. NVIDIA has clearly embraced a strategy of investing ahead of, and going after, growth markets for the chip business, and that strategy has paid off well. Thus, its current dominant positioning in the AI chip business can be viewed as more evidence of that strategy at play.

    There is one final component to NVIDIA's business model that needs noting, both from a profitability and risk perspective. NVIDIA 's core business is built around research and chip design, not chip manufacturing, and it outsources almost all of its chip production to TSMC. Its margins then come from its capacity to mark up the prices of these chips and it is exposed to the risks that any future China-Taiwan tensions can disrupt its supply chain.

2. Large, albeit Productive Reinvestment

    While NVIDIA's growth and profitability have been impressive, the value cycle is not complete until you bring in the investment that the company has  had to make to deliver that growth. With a semiconductor company, that reinvestment includes not only investing in manufacturing capacity, but also in the R&D to create the next generation of chips, in terms of power and capability. As with the sector, I capitalized R&D at NVIDIA, using a 5-year life, and recalculated my operating income (since the reported version is built on the accounting mis-reading of R&D as an operating expense). That results in a corrected version of pre-tax operating margin for NVIDIA that was 37.83% and a pre-tax return on capital of 24.42% in 2021-2023:

I also computed a sales to capital ratio, measuring the dollars of sales for each dollar of capital invested. In 2022, that number, for NVIDIA, was 0.65, indicating that this is definitely not a capital-light business and that NVIDIA has invested heavily to get to where it is today, as a company.

3. With a Mega Market Payoff

    NVIDIA's success on the operating front has impressed financial markets, and its rise in market capitalization from its IPO days to a trillion-dollar value can be seen below:

I know that there are many who are regretting their lack of foresight, in not owning NVIDIA through its entire run, but recognize that this was not a smooth ride to the top. In fact, the company had near-death experiences, at least in market value term, in 2002 and 2008, losing more than 80% of its market value. That said, I owe my lucky run with NVIDIA to one of those downturns in 2018, when the company lost more than 50% of its market value, and it is a lesson that I hope will come through this chart. Even the biggest winners in the market have had periods when investors have turned intensely negative on their prospects, making them attractive as investments for value-focused investors.

AI: From Promise to Profits

    Since much of the run-up in NVIDIA in the last few months has come from talk about AI, it is worth taking a detour and examining why AI has become such a powerful market driver, and perhaps looking at the past for guidance on how it will play out for investors and businesses.

Revolutionary or Incremental Change?

    I am old enough to be both a believer and a skeptic on revolutionary changes in markets, having seen major disruptors play out both in my personal life and my portfolio, starting with personal computers in the 1980s, the dot-com/online revolution in the 1990s, followed by smartphones in the first decade of this century and social media in the last decade. What set these changes apart was that they not only affected wide swathes of businesses, some positively and some adversely, but that they also changed the ways that we live, work and interact. In parallel, we have also seen changes that are more incremental, and while significant in their capacity to create new businesses and disruption, don't quite qualify as revolutionary. I won't claim to have any special skills in being able to distinguish between the two (revolutionary versus incremental), but I have to keep trying, since failing to do so will result in my losing perspective and making investing mistakes. Thus, I was unable to share the belief that some seemed to have about the "Cloud" and "Metaverse" businesses being revolutionary, since I saw them more as more incremental than revolutionary change. 

    So, where does AI fall on this spectrum from revolutionary to incremental to minimalist change? A year ago, I would have put it in the incremental column, but ChatGPT has changed my perspective. That was not because ChatGPT was at the cutting edge of AI technology, which it is not, but because it made AI relatable to everyone. As I watched my wife, who teaches fifth grade, grapple with students using ChatGPT to do homework assignments. and with my own students asking ChatGPT questions about valuation that they would have asked me directly, the potential for AI to upend life and work is visible, though it is difficult to separate hype from reality.    

Business Effects

    If AI is revolutionary change and will be a key market driver for this decade, what does this mean for investors? Looking back at the revolutionary changes from the last four decades (PCs, dot-com/internet, smartphones and social media), there are some lessons that may have application to the AI business.

1. A Net Positive for Markets? Does revolutionary change help the overall economy and/or equity markets? The results from the last four decades is mixed. The PC-driven tech revolution of the 1980s coincided with a decade of high stock market returns, as did the dot-com boom in the next decade, but the first decade of this century was one of the worst in market history as stock prices flatlined. Stocks did well again over the last decade, with technology as the big winner, and over the four decades of change (1980-2022), the annual return on stocks has been marginally higher than in the five decades prior. 

   

   Historical Stock Returns for US

   
   Given equity market volatility, four decades is a short time period, and the most that we can discern from this data is that the technological changes have been a net positive, for markets, albeit with added volatility for investors.
2. With a few Big Winners and Lots of Wannabes and Losers: It is indisputable that each of the revolutionary changes of the last four decades has created winners within the space, but a few caveats have also emerged. The first is that these changes have given rise to businesses where there are a few big winners, with a few companies dominating the space, and we have seen this paradigm play out with software, online commerce, smartphones and social media. The second is that the early leaders in these businesses have often fallen to the wayside and not become the big winners. Finally, each of these businesses, successful though they have been in the aggregate, have seen more than their share of false starts and failures along the way. For investors, the lesson has to be that investing in revolutionary change, ahead of others in the market, does not translate into high returns, if you back the wrong players in the race, or more importantly, miss the big winners. It is true that at this very early stage of the AI game, the market has anointed NVIDIA and Microsoft as big winners, but it is entirely possible that a decade from now, we will be looking at different winners. At the stage of the hype cycle, it is also true that almost every company is trying to wear the AI mantle, just as every company in the 1990s aspired to have a dot-com presence and many companies claimed to have "user-intensive" platforms in the last one, As investors, separating the wheat from the chaff will only get more difficult in the coming months and years, and it is part of the learning process. To the argument that you could buy a portfolio of companies that will benefit from AI and make money from the few that succeed, past market experience suggests that this portfolio is more likely to be over than under priced.
3. With Disruption: The market is littered with the carcasses of what used to be successful businesses that have been disrupted by technological change. Investors in these disrupted companies not only lose money, as they get disrupted, but worse, invest even more in them, drawn by their "cheapness". This happened, just to provide two examples, with investors in the brick-and-mortar retail companies that were devastated by online retail, and with investors in the newspaper/traditional ad companies that were upended by online advertising. If AI succeeds in its promise, will there be businesses that are upended and disrupted? Of course, but we are in the hype phase, where much more will be promised than can be delivered, but the biggest targets will come into focus sooner rather than later.

The bottom line is that even if we all agree that AI will change the way businesses and individuals behave in future years, there is no low-risk path for investors to monetize this belief. 

Value Effects

    If history is any guide, we are in the hype phase of AI, where it is oversold as the solution to just about every problem known to man, and used to justify large price premiums for the companies in its orbit, without any attempt to quantify and back up these premiums. The primary argument that will be used by those selling these AI premiums is that there is too much uncertainty about how AI will affect numbers in the future, an argument that is at odds with paying numbers up front for those expectations. In short, if you are paying a high price for an AI effect in a company, it behooves you to put aside your aversion to making estimates, and use your judgment (and data) to arrive at the effect of AI on cashflows, growth and risk, and by extension, on value.

    In making these estimates, it does make sense to break down AI companies based upon what part of the AI ecosystem they inhabit, and I would suggest the following breakdown:

• Hardware and Infrastructure: Every major change over the last few decades has brought with it requirements in terms of hardware and infrastructure, and AI is no exception. As you will see in the next section, the AI effect on NVIDIA comes from the increased demand for AI-optimized computer chips, and as that market is expected to grow exponentially, the companies that can grab a large share of this market will benefit.  There are undoubtedly other investments in infrastructure that will be needed to make the AI promise a reality, and the companies that are on a pathway to delivering this infrastructure will gain, as a consequence.
  
• Software: AI hardware, by itself, has little value unless it is twinned with software that can take advantage of that computing power. This software can take multiple forms, from AI platforms, chatbots, deep learning algorithms (including image and voice recognition, as well as natural language processing) and machine learning, and while there is less form and more uncertainty to this part of the AI business, it potentially has much greater upside than hardware, precisely for the same reason.

  

  Source

  
  
• Data: Since AI requires immense amounts of data, there will be businesses that will gain value from collecting and processing data specifically for AI applications. Big data, used more as a buzzword than a business proposition, over the last decade may finally find its place in the value chain, when twinned with AI, but that pathway will not be linear or predictable. 
• Applications: For companies that are more consumers of AI than its purveyors, the promise of AI is that it will change the way they do business, with positive and negative implications. The biggest pluses of AI, at least as presented by its promoters, is that it will allow companies to reduce costs (primarily by replacing manual labor with AI-driven applications) and make them more efficient, and by extension, more profitable. Even if I concede the first claim (though I think that the AI replacements will be neither as efficient nor as cost-saving as promised),  I am even more wary of the second claim for a simple reason. If every company has AI, and AI reduces costs and increases efficiency as promised for all of them, it is far more likely that they will end up with lower prices for their products/services and not higher profits. At the risk of repeating one of my favorite sayings, "If everyone has it, no one does" and it is the basis for my argument that AI, if it succeeds, will make companies less profitable, in the aggregate. The other minus of AI is that if it delivers on even a portion of its promise of automating aspects of business, it will be damaging and perhaps even devastating for existing companies that derive their value currently from delivering these services for lucrative fees. In these businesses, AI will not just be a zero-sum game, but a negative-sum one.

On the specific questions of how AI will affect investing, in general, and active investing, in specific, I believe that if it is used as a tool, it can enrich valuation and investing, and I look forward to being able to develop valuation narratives and numbers, with its aid. For those who are active investors, individuals as well as institutions, I believe that AI will make a difficult game (delivering excess returns or alpha from investing) even more so. Any edge you have as an active investor will be more quickly replicated in an AI world, and to the extent that AI tools will be accessible and available to every investor, by itself, AI will not be a sustainable edge for any active investor. 

Social Effects

    Will AI make our lives easier or more difficult? More generally, will it make the world a better or worse place to inhabit? I know that there are some advocates of AI who paint a picture of goodness, where AI takes over the menial tasks that presumably cause us boredom  and brings an unbiased eye to data analysis that lead to better decisions. I know that there are others who see AI as an instrument that big companies will use to control minds and acquire power. With the experience of the big changes that have engulfed us in the last few decades still fresh, I would argue that they are both right. AI will be a plus is some occupations and aspects of our lives, just as it will create unintended and adverse consequences in others.

    There are some who believe that AI can be held in check and made to serve its more noble impulses, by restricting or regulating its development, but I am not as optimistic for many reasons. First, I believe that both regulators and legislators are woefully incapable of understanding the mechanics of AI, let alone pass sensible restrictions on its usage, and even if they do, their motives are not altruistic. Second, any regulation or law that is aimed at preventing AI's excesses will almost certainly set in motion unintended consequences, that at least in some cases will be worse than the problems that the regulation/law was supposed to hold in check. Third, having seen how badly regulators and legislators have handled the consequences of the social media explosion, I am skeptical that they will even know where to start with AI. While this is a pessimistic take, I believe that it a realistic one, and that just as with social media, it will be up to us, as consumers of AI products and services, to try to draw lines and separate good from bad. We may not succeed, but what choice do we have, but to try?

The AI Chip Story

    The AI story has particular resonance with NVIDIA because unlike most other companies, where it is mostly hand-waving about potential, it has substance in place already and a market that is its target. In particular, NVIDIA has spent much of the last few years investing and developing products for a nascent AI market. This lead time has given NVIDIA not just market leadership, but revenues and profits already. Much of the excited reaction to NVIDIA's most recent earnings report came from the company reporting a surge in its data center revenues, with much of the increase coming from AI chips. While the company does not explicitly break out how much of the data center revenues are from AI chips, it is estimated that the total market for those chips in 2022 was about $15 billion, with NVIDIA holding a dominant market share of about 80%. If those estimates are right, the bulk of the data center revenues for NVIDIA in 2022, which amounted to $15 billion in all, comes from AI-optimized chips.

    The ChatGPT jolt to market expectations has played out in increases in expected growth of the AI chip market over the next decade, with estimates for the overall AI chip market in 2030 ranging from $200 billion at the low end to close to $300 billion at the high end. While there is a huge amount of uncertainty about this estimate, there are two assertions that can be made about NVIDIA's presence in this business. The first is that this will be the growth engine for NVIDIA's revenues over the next decade, even as their gaming and other chip revenue growth levels off. The second is that NVIDIA has a lead over its competition, and while AMD, Intel and TSMC will all allocate resources to building their AI businesses, NVIDIA's dominance will not crack easily.

NVIDIA: Valuation and Decision Time

    As you look at NVIDIA's growth and success in the last decade, and its recent ascent into the rarefied air of "trillion dollar market cap" companies, there are two impulses that come into play. One is to extrapolate the past and assume that assume that the company will continue to not just succeed in the future, but do so in a way that beats the market's expectations for it. The other is to argue that the outsized success of the past has raised investors expectations so much that it will be difficult for the company to meet them. In my story, I will draw on both impulses, and try to thread the needle on the company.

Story and Valuation

        The driver of NVIDIA's success has been its high-performance GPU cards, but it is very likely that the businesses that bought these cards and drove NVIDIA's success in the last decade will be different from the businesses that will make it successful in the next one. For much of the last decade, it was gaming and crypto users that allowed the company to set itself apart from the competition, but the bad news is that both of these markets are maturing, with lower expected growth in the future. The good news, for NVIDIA, is that it has two other businesses that are ready to step in and contribute to growth. The first is AI, where NVIDIA commands a hefty market share of what is now a relatively small market, but one that is almost certain to grow ten-fold or greater over the decade. The other is in the automobiles business, where more powerful computing is seen as the ingredient needed to open up automated driving and other enhancements. NVIDIA is only a small player in this space, and while it does not enjoy the dominance that it does in AI, a growing market will allow NVIDIA to acquire a significant market share. 

    I will start with a familiar construct (at least to those who follow my valuations), and break down the inputs that drive value as a precursor to introducing my NVIDIA story:

Put simply, the value of a company is a function of four broad inputs - revenue growth, as a stand-in for its growth potential, a target operating margin as a proxy for profitability, a reinvestment scalar (I use sales to invested capital) as a measure of the efficiency with which it delivers growth and a cost of capital & failure rate to incorporate risk. 

    While all of NVIDIA's different businesses (AI, Auto, Gaming) share some common features in terms of gross and operating margins, and requiring R&D for innovation, the businesses are diverging in terms of revenue growth potential. 

• Revenue Growth: NVIDIA will remain a high growth company for two reasons. The first is that in spite of its scaling up due to growth over the last decade, at least in terms of revenues, it has a modest market share of the overall semiconductor market, with revenues that are less than half of the revenues posted by Intel or TSMC. The second, and more important reason, is that while its gaming revenue growth is starting to flag, it is well-positioned in AI and Auto, two markets poised for rapid growth. In my story, I will assume that these markets will deliver on their growth promise and that NVIDIA will maintain a dominant, albeit lower, market share of the AI chip business, while gaining a significant share (15%) of the Auto chip business:
  
  Clearly, there is room for disagreement on both total market and market share for the AI and Auto businesses, and I will return to address the effects. I am still allowing the gaming and other business revenues to grow at 15% a year, a healthy number that reflects other businesses (like the omniverse) contributing to the top line.
• Profitability: The semiconductor business has a cost structure that has relatively little flex to it, but I will assume in my NVIDIA story that the right margin to focus on is the R&D adjusted version, and that NVIDIA will bounce back quickly from its 2022 margin setback to deliver higher margins than its peer group. While my target R&D adjusted margin of 40% may look high, it is worth remembering that the company delivered 42.5% as margin in 2020 and 38.4% as margin in 2021.  As noted earlier, NVIDIA's dependence on TSMC for the production of the chips it sells implies that any increases in margins have to come more from price increases than cost efficiencies.
• Investment Efficiency: NVIDIA has invested heavily in the last decade, generating only 65 cents in revenues for every dollar of capital invested (including the investment in R&D), in 2022. That investment has clearly been productive, as the company has been able to find growth and generate excess returns. I believe that given the company's larger scale, with the payoff from past investments augmenting revenues, the company's sales to invested capital will approach the global industry median, which is $1.15 in revenues for every dollar of capital invested.
• Risk: As we noted in the section on the semiconductor business, this remains, even for its most successful proponents, a cyclical business, and that cyclicality contributes to keeping the cost of capital higher than for the median company. I estimated NVIDIA's cost of capital based upon its geographic exposure and very low debt ratio to be 13.13%, but chose to use the industry average for US semiconductor companies, which was 12.21%, as the cost of capital in the initial growth period. Over time, I will assume that this cost of capital will drift down towards the overall market average cost of capital of 8.85%.

With this story in place, and the resulting input numbers, the value that I get for NVIDIA is shown below:

Download spreadsheet

Based on story, the value per share that I arrive at for NVIDIA on June 10, 2023, is about $240, well below the stock price of $409 that the stock traded at on June 10, 2023. (The stock has risen since then to $434 a share on June 20, 2023.)

Simulation and Breakeven Analysis

    At the risk of stating the obvious, I am making assumptions about market growth and market share that you may or even should take issue with. In the interests of examining how value varies as a function of the assumptions, I fell back on an approach that I find helps me deal with estimation uncertainty, which is a simulation. I built the simulation around the key inputs, including:

1. Revenues: In my base case valuation, incorporating high growth in the AI and Auto Chip businesses, and giving NVIDIA a dominant share of the first and a significant share of the second resulted in revenues of $267 billion in 2033. However, this is built on assumptions about the future for both markets that can be wrong, in either direction, and that uncertainty is incorporated into the simulation as distributions for each of the three segments of NVIDIA's revenues:
   
   As these distributions play out, there are simulations where NVIDIA's revenues exceed $600 billion and some where it is less than $100 billion, in 2033.
2. Operating Margin: In my base case story, I increase NVIDIA's R&D adjusted margin to 35% next year, and target an operating margin of 40% in 2027, that it maintains in perpetuity after that. While I provide my justifications for those assumptions, it is entirely possible that I am being too optimistic, in raising margins that are already above industry-average levels to even higher values, or that I am being pessimistic, and not factoring in NVIDIA's higher pricing power in the AI and Auto businesses. I capture that uncertainty in my (triangular) distribution for the target operating margin in 2027 (and beyond), where I set the upper end of the range at 50%, which would be a significant premium over NVIDIA's own past margins, and the lower end at 30%, which would put them closer to their peer group.
3. Reinvestment: The input that drives reinvestment is the sales to capital ratio, and while I set NVIDIA's sales to capital ratio at 1.15, the semiconductor industry average, it is possible that the company may continue to reinvest at closer to its historic average of 0.65 (leading to more reinvestment). Alternatively, it is also conceivable that the company's investments over the last decade, especially in its AI chips, will put it on a glide path to reinvesting a lot less in the next decade (a sales to capital ratio closer to 1.94, the 75th percentile of the semiconductor business.
4. Risk: Ruling out failure risk, and focusing on the cost of capital, I center my estimates on 12.21%, the industry average that I used in the base case, but allow for the possibility that a growing AI business may reduce the cyclicality of revenues, lowering the cost of capital towards the market-average of 8.85%) or conversely, increase uncertainty and uncertainty, raising the cost of capital towards 15%, the 90th percentile of global companies):

With these estimates in place, the simulated value per share is shown below:

To the question of whether NVIDIA could be worth $400 a share or more, the answer is yes, but the odds, at least based on my estimates, are low. In fact, the current stock price is pushing towards the 95th percentile of my value distribution.

    An alternative look at what has to happen for NVIDIA's intrinsic value to exceed $400, I looked at the two key variables that determine its value: revenues in year 10 and operating margins:

Download spreadsheet

This table reinforces the findings in the simulation, insofar as it shows that there are plausible paths that lead to the current price being a fair value or under value, but these paths require a daunting combination of extraordinary revenue growth and super-normal margins. In my view, a target margin of 50% is pushing the limits of possibility, in the semiconductor business, and if NVIDIA finds a way to deliver value that justifies current pricing, it has to be through explosive revenue growth. Put simply, you need another market or two, with potential similar to the AI market, where NVIDIA can wield a dominant market share to justify its pricing.

Judgment Day

    As I noted at the start of this post, I have a selfish reason for valuing NVIDIA, which is that I own it shares and I am exposed to its price movements, and much more so now than I was when I bought the stock in 2018, as a result of its inflated pricing. I have also been open about the fact that my investment philosophy is built around value, buying when price is less than value and by the same token, selling when price is much higher than value.

NVIDIA as an Investment

    I love NVIDIA as a company, and have nothing but praise for Jensen Huang's leadership of the company. Operating in a business where revenue growth was becoming scarce (single digit revenue growth) and segments of the product market are commoditized (lowering margins), NVIDIA found a pathway to not just deliver growth, but growth with superior profit margins and excess returns. While some may argue that NVIDIA was lucky to catch a growth spurt in the gaming and crypto businesses, a closer look at its successes suggests that it was not luck, but foresight, that put the company in a position to succeed. In fact, as the AI and Auto businesses look poised to grow, NVIDIA's positioning in both indicates that this is a company that is built to be opportunistic. My valuation story for NVIDIA reflects all of these positive features, and assumes that they will continue into the next decade, but that upbeat narrative still yields a value well below the current price.

    I would be lying if I said that selling one of my biggest winners is easy, especially since there is a plausible pathway, albeit a low-probability one, that the company will be able to deliver solid returns, at current prices. I chose a path that splits the difference, selling half of my holdings and cashing in on my profits, and holding on to the other half, more for the optionality (that the company will find other new markets to enter in the next decade). The value purists can argue, with justification, that I am acting inconsistently, given my value philosophy, but I am pragmatist, not a purist, and this works for me. It does open up an interesting question of whether you should continue to hold a stock in your portfolio that you would not buy at today's stock prices, and it is one that I will return to in a future post.

NVIDIA as a Trade

    I have written many posts about the divide between investing and trading, arguing that the two are philosophically different. In investing, you assess the value of a stock, compare that value to the price, act on that difference (buying when price is less than value and selling when it is greater) and hope to make money as the gap between value and price closes. In trading, you buy at a low price, hoping to sell at a higher price, but you are agnostic about what causes the price to move and whether that movement is rational or not. 

Bringing this difference to play in NVIDIA, you can see why, no matter what you think about NVIDIA's value, you may continue to trade it. Thus, even if you believe that NVIDIA's value is well below its price, you may buy NVIDIA on the expectation that the stock will continue to rise, borne upwards by momentum or incremental information. Given the strength of momentum as a market-driver, you may very well generate high returns over the next weeks, months or even years, and you should not let "value scolds" get in the way of your enjoyment of your winnings. My only pushback would be against those who argue that momentum can carry a stock forward forever, since it is the gift that both gives and takes away. The strength of momentum in the rise in NVIDIA's stock price will be played out in the the opposite direction, when (not if) momentum shifts, and if you are trading NVIDIA, you should be working on indicators that give you early warning of those shifts, not worrying about value.

The Bottom Line

    As we hear the relentless pitches for AI, and how it will change our live and affect our investments, there are lessons, to draw on, from the other big changes that we have seen over our lifetime. The first is that even if you buy into the argument that AI will change the ways that we work and play, it does not necessarily follow that investing in AI-related companies will yield returns. In other words, you can get the macro story right, but you need to also consider how that story plays out across companies to be able to generate returns. The second, is that refusing to make estimates or judgments about how AI will affect the fundamentals (cash flows, growth and risk) in a  business, just because you face significant uncertainty, will not make that uncertainty go away. Instead, it will create a vacuum that will be filled by arbitrary AI premiums and make us more exposed to scams and wannabes. The third is that, as a society, it is unclear whether adding AI to the mix will make us better or worse off, since every big technological change seems to bring with it unintended consequences. To end, I was considering asking ChatGPT to write this post for me, using my own language and history, and I am open to the possibility that it could do a better job than I have. Stay tuned!

YouTube

Spreadsheets

1. NVIDIA Valuation (June 2023)

Good (Bad) Banks and Good (Bad) Investments: At the right price...

In my last post, I looked at banking as a business, and used  a simple banking framework to advance the notion that the key ingredient tying together the banks that have failed so far in 2023 is an absence of stickiness in deposits, created partially by depositor and deposit characteristics (older are  stickier than younger) and partly by growth in deposits (high growth increases stickiness). I also used the banking framework to argue that good banks have stickier deposits, with a higher precent of these deposits being non-interest bearing, that they invest in loans and investment securities on which they earn interest rates that cover and exceed the default risk in these investments. While differentiating between good and bad banks can be straightforward, it does not follow that buying good banks and selling bad banks is a good investment strategy, since its success depends entirely on what the market is incorporating into stock prices. An investor who buys a good bank at too high a price, given its goodness, will underperform one who buys a bad bank at too low a price, given its badness. In this post, I will begin by looking at  how to value banks and follow up with an examination of investor views of banking have changed, by looking at pricing, before examining divergences in how banks are priced in the market today.

The Intrinsic Value of Bank Equity

   I am a dabbler in all things valuation-related, and I find the process fascinating, as stories about businesses get translated into valuation inputs, and finally into value. I enjoy challenging valuations, but banks remain, at least for me, the last frontier in valuation, simply because so much of what we do in conventional valuation does not work with banks, and a crisis or panic can upend even the most carefully done bank valuation.

All Equity, All the time!

    With most non-financial service businesses, you face a choice in how you approach valuation. You can value the enterprise or the entire business, focusing on valuing the operations or assets of the business, and consider capital as inclusive of both debt and equity. Alternatively, you can value just the equity in the business, focusing on cash flows left over after debt payments and discounting back at a rate of return that reflects the risk that equity investors face:

With banks, this choice does not exist, since debt to a bank can be expansively defined to include deposits as well, making it effectively raw material for the bank's operations, where the objective is borrow money (from depositors and lenders) at a low rate and lend it out or invest it at a higher rate. Consequently, you can only value the equity in a bank, and by extension, the only pricing multiples you can use to price banks are equity multiples (PE, Price to Book etc.).  The notion of computing a cost of capital for a bank is fanciful and fruitless, and any attempt to compute an enterprise value for a bank is destined to end in failure.

Equity Valuation 101 and Dividend Discount Model

    Staying on equity valuation, the intrinsic value of equity can be written as a function of the cash flows left for equity investors, after reinvestment and taxes, and after all other claim holders have been paid, and the cost of equity:

Over the decades, analysts trying to put this model into practice with banks have run into trouble estimating cash flows for banks, using the traditional structure, since items like capital expenditures and working capital are impossible to measure at banks. It should come as no surprise that, at least with banks, analysts fell back on the only observable cash flows to equity, i.e., dividends;

It is this line of reasoning, i.e., that it is difficult, if not impossible, to estimate banking free cashflows to equity, that I used prior to 2008, when I argued for the use of the dividend discount model to value banks In using the dividend discount model, I was making two implicit assumptions. The first was that banks were run by sensible people, who paid out what they could afford to in dividends, neither holding back on paying dividends nor paying too much in dividends. The other was that the bank regulatory framework operated effectively, preventing banks from overreaching on risk or being under capitalized. 

A Bank FCFE Model

    The events of 2008 dispelled me of both delusions that allowed for the use of the dividend discount model, as it became clear that the managers of banks were anything but sensible and the regulatory framework had large holes in it that were exploited. In the years after, I have replaced dividends with a variant on free cash flow to equity, defined through the lens of a banking business, discounted back at a cost of equity reflecting banking risks (duration mismatches, low regulatory capital and riskiness of loan/investment portfolios).

Note the differences between the bank FCFE and bank dividend discount models. 

• The first is that unlike dividends, which are floored at zero, the free cash flow to equity for a growing or severely undercapitalized bank can be negative, reflecting the need to raise fresh equity to survive. 
• The other is that by tying cashflows to capitalization, it allows us to bring in that same factor into risk and costs of equity, with under capitalized banks have higher costs of equity. 
• As a final component of bank equity value, and 2023 has brought this home to us is the reality that even a healthy, profitable bank can see its value melt away in days, if its depositors decide, for good, bad or no reasons at all, to withdraw their deposits and put the bank into the death spiral from which recovery can be close to impossible. Since this risk is existential, it is almost impossible to build into a discounted cash flow model, which is for a going-concern, and has to be incorporated as a risk of failure.

In short, the banking version of a FCFE model gives us access to levers that allow us to differentiate across banks and bring in the elements that make some banks better than others.

Valuing Citi

    Intrinsic valuation models connect only when applied to real companies, and in the table below, I used the Bank FCFE model described above in my valuation of Citi (a choice that may strike you as odd, but which you will understand if you read the rest of my post). To set the table, in the battle of big banks for investor acclaim, Citi has clearly lost the battle not only against JP Morgan Chase, but against most of the other big US banks. It has delivered low growth and subpar profitability, but it has built up buffers in its capital ratios and still has a banking model that delivers a lucrative interest rate spread.

    In my valuation, I will assume that Citi will continue on its cautious, low-growth path, growing its risk-adjusted assets at 3% a year in perpetuity, a little lower than its 3.74% growth rate over the last 5 years. Over time, I expect some improvement in the return on equity, which was 8.78% in 2022, to its five-year average of 9.50%, which is still lower than the cost of equity of 11.67% that I am using for large commercial banks (see picture below for how I am computing an implied cost of equity for the 25 largest banks). Finally, I will assume that the bank will continue to marginally improve its Tier 1 capital ratio, currently at 14.80% to reach a target of 15.00%, in five years:

Download spreadsheet to value Citi (or any bank)

Note that the combination of low growth and a healthy, current regulatory capital ratio keep the needs for reinvesting in regulatory capital low, allowing for large potential dividends. Those high cash flows, even though they are delivered by a bank that earns and expects to continue to earn an ROE less than its cost of equity translate into a value of equity for Citi of about $69, making it about 32% under valued auto the stock price of $46.32, at close of trading on May 5, 2023. Obviously, you will have very different views about Citi than I do, and you are welcome to download the spreadsheet and input your numbers not just for Citi, but for any bank.

The Pricing of Bank Equity

    You have heard me say this before, but I don't think there is any harm in repeating this. Value and price are words that are often used interchangeably, but they come from different processes and can yield different numbers for the same asset or company. 

Since pricing requires comparison across companies, often with different units (numbers of shares outstanding), we generally convert market values into pricing multiples, to allow for this comparison. As we noted in the last section, the pricing multiples that we use to compare banks have to be equity multiples, with price earnings rations and price to book ratios being the most common choices. 

Price to Book Ratio: Choice and Drivers

   There is no sector where price to book ratios get used more than in banking and financial services, for two reasons. The first is that the book value of equity for a bank, by becoming the basis for regulatory capital, has operating consequences, since actions or events that lower than book value of equity (an unexpected loss, a regulatory fine, a trading shortfall) can cause a bank to become undercapitalized and go out of business. The other is that marking to market is more common in banks than at other sectors, the hold-to-maturity loophole notwithstanding, and that should make book value of equity a more meaningful figure than book value of equity at a software or a consumer product company. 

    To use the price to book ratio to price banks, I begin by identifying its drivers, and that is simple to do, if you start with an intrinsic equity valuation model. In fact, using the simplest equity valuation model that I can think of, which is a stable growth dividend discount model, we get:

Dividing both sides by book value of equity, and setting growth rate = (1- Payout ratio) ROE, we can simplify this equation:

This equation, in its simple, stable growth form, suggests that whether a bank trades at below or above its book value of equity will be driven by whether investors expect banks to earn more than their cost of equity (price to book>1), roughly the cost of equity (price to book = 1) or less than the cost of equity (price to book <1).

Price to Book for Banks: Over time

    To get a measure of how banks are being priced today, it is worth getting perspective as to how investor views on bank profitability and risk have changed over time. To gain this perspective, I looked at the aggregated price to book ratio of all US banks, obtained by first aggregating the market capitalizations of all banks and dividing by the aggregated book equity from 2004 to 2022, at the end of each year, and in May 2023:

    

If there is a lesson in the graph, it is that the 2008 crisis has left a lasting impression, as US banks have struggled since that crisis to elevate price to book ratios. Even as returns on equity have slowly recovered close to pre-2008 levels, the price to book ratios have not recovered, even as the rest of the market has seen rising price to book ratios, due to lower interest rates. In fact, the 2023 crisis has reduced the aggregate price to book ratio for US banks to close to one, the Maginot line below which investors are assuming that banks will generate return on equity roughly equal to their cost of equity in the long term.

    For some, this drop in price to book ratios over time is a sign of market overreaction, and there are some value investors who have overweighted their bank holdings as a consequence. That may very well be the case, but I think it is prudent to see if there are fundamental reasons for the shift:

• Higher Risk: One explanation is that investors perceive banks to be riskier than they were prior to 2008, and at first sight, that seems puzzling since banks have become better capitalized over the last decade, as the regulatory authorities have reacted to the 2008 crisis by tightening safety capital requirements: 
  
  
  
  As you can see the Tier 1 capital at US banks collectively has risen to 13-14% from 10-11% in the years leading into the 2008 crisis and after. It is true, though, that equity as a percent of total assets dipped especially in 2020 and 2021, before bouncing back, but even that statistic has shown little change over the decade.
• Lower Profitability: Another is that investors don't trust net income reported by banks as final numbers, given the propensity of some banks to surprise them with after-the fact and unexpected losses (from trading mistakes and asset write-downs) or believe that banks are becoming less profitable over time. To see if this is the case, I looked at the interest income and expenses over time at banks:
  
  
  
  Like the book equity, the spread dropped in 2020 and 2021, with 2022 showing a recovery. However, as interest rates have risen, it is likely that rates on deposits will rise faster than rates earned on loans and investments in the near term, perhaps a source of concern for investors.
• Business Economics; If banks are not more risky, at least collectively, and interest spreads have held their own, the only remaining explanation is that investors believe that the banking business is much less likely to be value-adding now than two decades ago. Almost every aspect of banking is under stress, with deposits becoming less sticky, increased competition for the loan business from fintech and other disruptors and increased risks of contagion and crisis, and while banks remain profitable on the surface, investors have good reason to wonder whether the good times will last.

I know that a contrarian take, especially after the last few months, would suggest investing in banking as a sector, but I believe that the long term trends for the business are negative.

The Pricing of Banks - Across Banks

    Even though I would not make a collective bet on banks collectively, I do believe that, as in any crisis, individual banks are getting mis-placed. Thus, as investors panic and sell regional banks, it is likely that good regional banks and lumped in bad ones, in the sell off,  and if that money is being redirected to the bigger banks, some of those banks may not merit the price increases. Staying with price to book as my pricing metric for banks, I looked at the distribution of price to book ratios across banks, both at the end of 2022, and in May 2023, as the banking crisis has unfolded:

As you can see, the crisis has lowered price to book ratios across the board, with the median price to book ratio dropping from 1.12 at the end of 2022 to 0.94 in May 2022. That decline is almost entirely the result of a decline in market capitalization, since the book values of equity for banks were little changed between the third quarter of 2022 (used for the end of 2022 calculation) and the end of 2022 (used for the May 2023 calculation).

    Since the key driver of price to book ratio is the return on equity, I looked at the distribution of returns on equity at US banks in 2021 and 2022:

Unlike companies in other sectors, where there are wide variations across companies, the returns on equity at banks is tightly clustered, with 50% of banks having 9.38% (9.24%) and 14.80% (13.75%) in 2021 (2022). However, there are clearly banks that generate higher returns on equity than other banks, and that should play a role in explaining differences in price to book ratios. To check how closely price to book ratios at banks hew to the returns on equities generated by banks, I did a scatter plot of price to book against ROE, both at the end of 2022 and again in May 2023:

While banks with higher ROE generally trade at higher price to book ratios, there is significant noise in the relationship, though more in May 2023 (with an R squared just above 11%) than in December 2022 (with an R squared just above 14%). In a final visual display, I looked a 3D scatter plot, of PBV against ROE and Tier 1 capital ratios:

I have highlighted the combination that characterizes the most under valued banks (low price to book, high ROE and a high Tier 1 capital ratio) as well as the combination for the most over valued banks (high price to book, low ROE and low Tier 1 capital

The Biggest Banks- Trawling for Bargains!

    One of the exercises that I find useful, when pricing,  is to look for the perfect underpriced stock, one that looks cheap with no good reason for why it is so cheap. Applying that practice to banks, here is what you would want to see in your underpriced bank:

Applying this approach to the 25 largest banks, for instance, I computed the median values for each of these variables for the 25 largest US banks, in terms of market cap, and used it as the dividing line for good and bad on each of the variables. Thus, a return on equity higher than the median of 12% is considered a good (and in green) and less than 12% is considered bad (and in red). 

Source: S&P Capital IQ

Put simply, you are looking for a preponderance of green numbers for your under priced banks, and while there no all green banks, Citi comes closest to meeting the tests,  scoring well on risk (with a higher than median Tier 1 capital ratio and a lower percent of securities held to maturity among the five biggest banks), deposit stickiness (with low deposit growth) and  trades at half of book value (the lowest price to book ratio). Its weakest link is a return on equity of 8.11% (in 2022) and 9.50% (average from 2018-2022), lower than the median for US banks, and while that would suggest a lower than median price to book ratio, the discount at Citi exceeds that expectation. Citi's banking business, though slow growing, remains lucrative with the higher interest rate spread in this sample. I will be adding Citi to my portfolio, since it offers the best mix of cheapness and low risk, and hope that it can at least maintain its profitability, though an increase would be icing on the cake. It is a slow-growth, stodgy bank that seems to be priced on the presumption that it will not only never earn a ROE even close to its cost of equity, and that makes it a good investment. 

    At the other end of the expectation scales, JP Morgan Chase scores well on operating metrics, with a high ROE, low deposit growth and a high Tier 1 capital ratio, but it trades at a much higher price to book ratio than the other banks, and iwith a lower dividend yield. I have owned JPM Chase for close to a decade in my portfolio, and I don't see anything in this table that would lead me to sell, though I would not be in a hurry to buy either, at today's prices, if I did not own it.

    As a value investor, I would be uncomfortable making an investment in Citi, purely based upon this pricing analysis, and it is for that reason that I retraced my steps to do the intrinsic valuation of the bank that you saw in the last section. That reinforces a more general point that even investors who are true believers in valuation can benefit from understanding and using pricing, just as traders, who play the pricing game, can benefit from an understanding of the core principles of intrinsic valuation.

Wrapping up

In my last post, my focus was on the qualities that made for the distinction between good and bad banks, and those qualities includes deposit stickiness, a low interest rate on deposits combined with a high interest rate (given default risk) on loans and investments and a big buffer against shocks (with high book equity and Tier 1 capital ratios). In this post, I  shifted attention to the investing side of the picture, and that puts the price you pay to acquire banks on center stage. Acquiring a good bank, while paying too high a price, will make for a bad investment, just as acquiring a bad bank, at a bargain price, will be a good investment. On the contest of banking quality, JP Morgan Chase would beat Citi handily, with a high return on equity and continued growth, combined with safety, but in the contest for investing dollars, Citi is the better priced bank. Since I will have both stocks in my portfolio starting tomorrow, I will have a ringside seat to watch this contest play out over the next few years.

YouTube Video

Posts on the Banking Crisis 2023

1. Breach of Trust: Decoding the Banking Crisis (of 2023)
2. Good (Bad) Banks and Good (Bad) Investments: At the right price...

Bank Valuation Spreadsheet

1. Bank Valuation Spreadsheet (with Citi Valuation)

Breach of Trust: Decoding the Banking Crisis

In March 2023, the fall of Silicon Valley Bank shocked investors not only because it was unforeseen, but also because of the speed with which it unfolded. That failure has had a domino effect, with Signature Bank falling soon after, followed by Credit Suisse in April 2023 and by First Republic last week. The banks that have fallen so far collectively controlled more deposits than all of the banks that failed in 2008, but unlike that period, equity markets in the United States have stayed resilient, and even within banking, the damage has varied widely across different segments, with regional banks seeing significant draw downs in deposits and market capitalization. The overarching questions for us all are whether this crisis will spread to the rest of the economy and market, as it did in 2008, and how banking as a business, at least in the US, will be reshaped by this crisis, and while I am more a dabbler than an expert in banking, I am going to try answering those questions.

The Value of a Bank

    Banks have been an integral part of business for centuries, and while we have benefited from their presence, we have also been periodically put at risk, when banks over reach or get into trouble, with their capacity to create costs that the rest of us have to bear. After every banking crisis, new rules are put into place to reduce or minimize these risks to the economic system, but in spite of these rules or sometimes because of them, there are new crisis. To understand the roots of bank troubles, it is important that we understand how the banking business works, with the intent of creating criteria that we can use to separate good banks from average or bad ones.

The Banking Business Model

    The banking business, when stripped down to basics, is a simple one. A bank collects deposits from customers, offering the quid quo pro of convenience, safety and sometimes interest income (on those deposits that are interest-beating) and either lends this money out to borrowers (individuals and businesses), charging an interest rate that is high enough to cover defaults and leave a surplus profit for the bank. In addition, banks can also invest some of the cash in securities, usually fixed-income, and with varying maturities and degrees of default risk, again earning income from these holdings. The profitability of a bank rests on the spread between its interest income (from loans and financial investments) and its interest expenses (on deposits and debt), and the leakages from that spread to cover defaulted loans and losses on investment securities:

To ensure that a bank survives, it's owners have to hold enough equity capital to buffer against unanticipated defaults or losses. 

The Bank Regulators

    If you are wondering where bank regulators enter the business model, it is worth remembering that banks predate regulators, and for centuries, were self regulated, i.e., were responsible for ensuring that they had enough equity capital to cover unexpected losses. Predictably, bank runs were frequent and the banks that survived and prospered set themselves apart from the others by being better capitalized and  better assessors of default risk than their competition. In the US, it was during the civil war that the National Banking Act was passed, laying the groundwork for chartering banks and requiring them to maintain safety reserves. After a banking panic in 1907, where it fell upon J.P. Morgan and other wealthy bankers to step in and save the system, the Federal Reserve Bank was created in 1913. The Great Depression gave rise to the Glass-Steagall Act in 1933 which restricted banks to commercial banking, with the intent of preventing them from investing their deposit money in riskier businesses. The notion of regulatory capital has always been part of bank regulation, with the FDIC defining "capital adequacy" as having enough equity capital to cover one-tenth of assets. In subsequent decades, these capital adequacy ratios were refined to allow for risk variations across banks, with the logic that riskier assets needed more capital backing than safer ones. These regulatory capital needs were formalized and globalized after the G-10 countries created the Basel Committee on Banking Supervision and explicitly created the notions of "risk-weighted assets" and "Tier 1 capital", composed of equity and equity-like instruments, as well as specify minimum capital ratios that banks had to meet to continue to operate. Regulators were given punitive powers, ranging from restrictions of executive pay and acquisitions at banks that fell below the highest capitalization ranks to putting banks that were undercapitalized into receivership.

    The Basel accord and the new rules on regulatory capital have largely shaped banking for the last few decades, and while they have provided a safety net for depositors, they have also given rise to a dangerous game, where some banks arrived at the distorted conclusion that their end game was exploiting loopholes in regulatory capital rules, rather than build solid banking businesses. In short, these banks found ways of investing in risky assets that the regulators did not recognize as risky, either because they were new or came in complex packages, and using non-equity capital (debt and deposits), while getting that capital classified as equity or equity-like for regulatory purposes. The 2008 crisis  exposed the ubiquity and consequences of this regulatory capital game, but at great cost to the economy and tax payers, with the troubled assets relief program (TARP) investing $426 billion in bank stocks and mortgage-backed securities to prop up banks that had over reached, mostly big, money-center banks, rather than small or regional banks. The phrase "too big to fail" has been over used, but it was the rationale behind TARP and is perhaps at the heart of today's banking crisis.

Good and Bad Banks

    If the banking business is a simple one, what is that separates good from bad banks? If you look back at the picture of the banking business, you can see that I have highlighted key metrics at banks that can help gauge not just current risk but their exposure to future risk. 

1. Deposits: Every banks is built around a deposit base, and there are deposit base characteristics that clearly determine risk exposure. First, to the extent that some deposits are not interest-bearing (as is the case with most checking accounts), banks that have higher percentages of non-interest bearing deposits start off at an advantage, lowering the average interest rate paid on deposits. Second, since a big deposit base can very quickly become a small deposit base, if depositors flee, having a stickier deposit base gives a bank a benefit. As to the determinants of this stickiness, there are numerous factors that come into play including deposit size (bigger and wealthier depositors tend to be more sensitive to risk whispers and to interest rate differences than smaller ones), depositor homogeneity (more diverse depositor bases tend to be less likely to indulge in group-think) and deposit age (depositors who have been with a bank longer are more sticky). In addition to these bank-specific characteristics, there are two other forces that are shaping deposit stickiness in 2023. One is that the actions taken to protect the largest banks after 2008 have also tilted the scales of stickiness towards them, since the perception, fair or unfair, among depositors, that your deposits are safer at a Chase or Citi than they are at a regional bank. The other is the rise of social media and online news  made deposits less sticky, across the board, since rumors (based on truth or otherwise) can spread much, much faster now than a few decades ago.
2. Equity and Regulatory Capital: Banks that have more book equity and Tier 1 capital have built bigger buffers against shocks than banks without those buffers. Within banks that have high accumulated high amounts of regulatory capital, I would argue that banks that get all or the bulk of that capital from equity are safer than those that have created equity-like instruments that get counted as equity. 
3. Loans: While your first instinct on bank loans is to look for banks that have lent to safer borrowers (less default risk), it is not necessarily the right call, when it comes to measuring bank quality. A bank that lends to safe borrowers, but charges them too low a rate, even given their safer status, is undercutting its value, whereas a bank that lends to riskier borrowers, but charges them a rate that incorporates that risk and more, is creating value. In short, to assess the quality of a bank's loan portfolio, you need to consider the interest rate earned on loans in conjunction with the expected loan losses on that loan portfolio, with  a combination of high  (low) interest rates on loans and low (high) loan losses characterizing good (bad) banks. In addition, banks that lend to a more diverse set of clients (small and large, across different business) are less exposed to risk than banks that lend to homogeneous clients (similar profiles or operate in the same business), since default troubles often show up in clusters.
4. Investment Securities: In the aftermath of the 2008 crisis, where banks were burned by their holdings in riskier mortgage-backed securities, regulators pushed for more safety in investment securities held by banks, with safety defined around default and liquidity risk. While that push was merited, and banks with safer and more liquid holdings are safer than banks with riskier, illiquid  holdings, there are two other components that also determine risk exposure. The first is the duration of these securities, relative to the duration of the deposit base, with a greater mismatch associated with more risk. A bank that is funded primarily with demand deposits, which invests in 10-year bonds, is exposed to more risk than if invests in commercial paper or treasury bills. The second is whether these securities, as reported on the balance sheet, are marked to market or not, a choice determined (at least currently) by how banks classify these holdings, with assets held to maturity being left at original cost and assets held for trading, being marked to market. As an investor, you have more transparency about the value of what a company holds and, by extension, its equity and Tier 1 capital, when securities are marked to market, as opposed to when they are not.

At the risk of over simplifying the discussion, the picture below draws a  contrast between good and bad banks, based upon the discussion above:

Banks with sticky deposits, on which they pay low interest rates (because a high percentage are non-interest bearing) and big buffers on equity and Tier 1 capital, which also earn "fair interest rates", given default risk, on the loans and investments they make, add more value and are usually safer than banks with depositor bases that are sensitive to risk perceptions and interest rates paid, while earning less than they should on loans and investments, given their default risk. 

Macro Stressors

    While we can differentiate between good and bad banks, and some of these differences are driven by choices banks make on how they build their deposit bases and the loans and investments that they make with that deposit money, these differences are often either ignored or overlooked in the good times by investors and regulators. If often requires a crisis for both groups to wake up and respond, and these crises are usually macro-driven:

1. Recessions: Through banking history, it is the economy that has been the biggest stressor of the banking system, since recessions increase default across the board, but more so at the most default-prone borrowers and investment securities. Since regulatory capital requirements were created in response to one of the most severe recessions in history (the Great Depression), it is not surprising that regulatory capital rules are perhaps most effective in dealing with this stress test.
2. Overvalued Asset Classes: While banks should lend money using a borrower's earnings capacity as collateral, it is a reality that many bankers lend against the value of  assets, rather than their earning power. The defense that bankers offer is that these assets can be sold, if borrowers default, and the proceeds used to cover the outstanding dues. That logic breaks down when asset classes get overvalued, since the loans made against the assets can no longer be covered by selling these assets, if prices correct. This boom and bust cycle has long characterized lending in real estate, but became the basis for the 2008 crisis, as housing prices plunged around the country, taking down not just lenders but also holders of real-estate based securities. In short, when these corrections happen, no matter what the asset class involved, banks that are over exposed to that asset class will take bigger losses, and perhaps risk failure.
3. Inflation and Interest Rates: Rising inflation and interest rates are a mixed blessing for banks. On the one hand, as rates rise, longer life loans and longer term securities will become less valuable, causing losses.. After all, the market price of even a default-free bond will change, when interest rates change, and bonds that were acquired when interest rates were lower will become less valuable, as interest rates rise. In most years, those changes in rates, at least in developed markets like the US, are small enough that they create little damage, but 2022 was an uncommon year, as the treasury bond rate rose from 1.51% to 3.88%, causing the price of a ten-year treasury bond to drop by more than 19%.

Put simply, every bank holding ten-year treasury bonds in 2022 would have seen a mark down of 19% in the value of these holdings during the year, but as investors, you would have seen the decline in value only at those few banks which classified these holdings as held for sale. That pain becomes worse with bonds with default-risk, with Baa (investment grade) corporate bonds losing 27% of their value. On the other hand, banks that have higher percentages of non-interest bearing deposits will gain value from accessing these interest-free deposits in a high interest world. The net effect will determine how rising rates play out in bank value, and may explain why the damage from the crisis has varied across US banks in 2023.

The Banks in Crisis

    It is worth noting that all of the pain that was coming from writing down investment security holdings at banks, from the surge in interest rates, was clearly visible at the start of 2023, but there was no talk of a banking crisis. The implicit belief was that banks would be able to gradually realize or at least recognize these losses on the books, and use the time to fix the resulting drop in their equity and regulatory capital. That presumption that time was an ally was challenged by the implosion of Silicon Valley Bank in March 2023, where over the course of a week, a large bank effectively was wiped out of existence. To see why Silicon Valley Bank (SVB)  was particularly exposed, let us go back and look at it through the lens of good/bad banks from the last section:

1. An Extraordinary Sensitive Deposit Base: SVB was a bank designed for Silicon Valley (founders, VCs, employees) and it succeeded in that mission, with deposits almost doubling in 2021. That success created a deposit base that was anything but sticky, sensitive to rumors of trouble, with virally connected depositors drawn from a common pool and big depositors who were well positioned to move money quickly to other institutions. 
2. Equity and Tier 1 capital that was overstated: While SVB's equity and Tier 1 capital looked robust at the start of 2023, that look was deceptive, since it did not reflect the write-down in investment securities that was looming. While it shared this problem with other banks, SVB's exposure was greater than most (see below for why) and explains its attempt to raise fresh equity to cover the impending shortfall.
3. Loans: A large chunk of SVB's loan portfolio was composed of venture debt, i.e., lending to pre-revenue and money-losing firms, and backed up by expectations of cash inflows from future rounds of VC capital. Since the expected VC rounds are conditional on these young companies being repriced at higher and higher prices over time, venture debt is extraordinarily sensitive to the pricing of young companies. In 2022, risk capital pulled back from markets and as venture capital investments dried up, and down rounds proliferated, venture debt suffered. 
4. Investment Securities: All banks put some of their money in investment securities, but SVB was an outlier in terms of how much of its assets (55-60%) were invested in treasury bonds and mortgage-backed securities. Part of the reason was the surge in deposits in 2021, as venture capitalists pulled back from investing and parked their money in SVB, and with little demand for venture debt, SVB had no choice but to invest in securities. That said, the choice to invest in long term securities was one that was made consciously by SVB, and driven by the interest rate environment in 2021 and early 2022, where short term rates were close to zero and long term rates were low (1.5-2%), but still higher than what SVB was paying its depositors. If there is an original sin in this story, it is in this duration mismatch, and it is this mismatch that caused SVB's fall.

In short, if you were building a bank that would be susceptible to a blow-up, from rising rates, SVB would fit the bill, but its failure opened the door for investors and depositors to reassess risk at banks at precisely the time when most banks did not want that reassessment done.

    In the aftermath of SVB's failure, Signature Bank was shut down in the weeks after and First Republic has followed, and the question of what these banks shared in common is one that has to be answered, not just for intellectual curiosity, because that answer will tell us whether other banks will follow. It should be noted that neither of these banks were as exposed as SVB to the macro shocks of 2022, but the nature of banking crises is that as banks fall, each subsequent failure will be at a stronger bank than the one that failed before.  

• With Signature Bank, the trigger for failure was a run on deposits, since more than 90% of deposits at the bank were uninsured, making those depositors far more sensitive to rumors about risk. The FDIC, in shuttering the bank, also pointed to "poor management" and failure to heed regulatory concerns, which clearly indicate that the bank had been on the FDIC's watchlist for troubled banks.
• With First Republic bank, a bank that has a large and lucrative wealth management arm, it was a dependence on those wealthy clients that increased their exposure. Wealthy depositors not only are more likely to have deposits that exceed $250,000, technically the cap on deposit insurance, but also have access to information on alternatives and the tools to move money quickly. Thus, in the first quarter of 2023, the bank reported a 41% drop in deposits, triggering forced sale of investment securities, and the realization of losses on those sales.

In short, it is the stickiness of deposits that seems to be the biggest indicator of banks getting into trouble, rather than the composition of their loan portfolios or even the nature of their investment securities, though having a higher percentage invested in long term securities leaves you more exposed, given the interest rate environment. That does make this a much more challenging problem for banking regulators, since deposit stickiness is not part of the regulatory overlay, at least at the moment. One of the outcomes of this crisis may be that regulators monitor information on deposits that let them make this judgment, including:

1. Depositor Characteristics: As we noted earlier, depositor age and wealth can be factors that determine stickiness, with younger and wealthier depositors being less sticky that older and poorer depositors. At the risk of opening a Pandora's box, depositors with more social media presence (Twitter, Facebook, LinkedIn) will be more prone to move their deposits in response to news and rumors than depositors without that presence.
2. Deposit age: As in other businesses, a bank customer who has been a customer for longer is less likely to move his or her deposit, in response to fear, than one who became a customer recently. Perhaps, banks should follow subscriber/user based companies in creating deposit cohort tables, breaking deposits down based upon how long that customer has been with the bank, and the stickiness rate in each group. 
3. Deposit growth: In the SVB discussion, I noted that one reason that the bank was entrapped was because deposits almost doubled in 2021. Not only do very few banks have the capacity to double their loans, with due diligence on default risk, in a year, but these deposits, being recent and large, are also the least sticky deposits at the bank. In short, banks with faster growth in their deposit bases also are likely to have less sticky depositors.
4. Deposit concentration: To the extent that the deposits of a bank are concentrated in a geographic region, it is more exposed to deposit runs than one that has a more geographically diverse deposit base. That would make regional bank deposits more sensitive that national bank deposits, and sector-focused banks (no matter what the sector) more exposed to deposit runs than banks that lend across businesses. 

Some of this information is already collected at the bank level, but it may be time for bank regulators to work on measures of deposit stickiness that will then become part of the panel that they use to judge exposure to risk at banks.

The Market Reaction

    The most surprising feature of the 2023 banking crisis has been the reaction of US equity markets, which have been resilient, rising in the face of a wall of worry. To illustrate how the market reaction has played out at different levels, I looked at four indices, starting with the S&P 500, moving on the S&P Financials and Banks Index to the S&P Select Bank Index and finally, the S&P Regional Bank Index.

The S&P 500 index is up 6.5% this year, indicative of the resilience on the part of the market, or denial on the part of investors, depending on your perspective. The S&P Financial Sectors index is down 5.72%, but the S&P Select Banks index is down 26.2% and the regional bank index has taken a pummeling, down more than 35%. The damage from this banking crisis, in short, has been isolated to banks, and within banks, has been greater at regional banks than at the national banks.

    The conventional wisdom seems to bethat big banks have gained at the expense of smaller banks, but the data is more ambiguous. I looked at the 641 publicly traded US banks, broken down by market capitalization at the start of 2023 into ten deciles and looked at the change in aggregate market cap within each decile. 

As you can see the biggest percentage declines in market cap are bunched more towards the bigger banks, with the biggest drops occurring in the eighth and ninth deciles of banks, not the smallest banks. After all, the highest profile failures so far in 2023 have been SVB, Signature Bank and First Republic Bank, all banks of significant size. 

    If my hypothesis about deposit stickiness is right, it is banks with the least stick deposits that should have seen the biggest declines in market capitalization. My proxies for deposit stickiness are limited, given the data that I have access to, but I used deposit growth over the last five years (2017-2022) as my  measure of stickiness (with higher deposit growth translating into less stickiness):

The results are surprisingly decisive, with the biggest market capitalization losses, in percentage terms, in banks that have seen the most growth in deposits in the last five years. To the extent that this is correlated with bank size (smaller banks should be more likely to see deposit growth), it is by no means conclusive evidence, but it is consistent with the argument that the stickiness of deposits is the key to unlocking this crisis.

Implications

    I do believe that there are more dominos waiting to fall in the US banking business, with banks that have grown the most in the last few years at the most risk, but I also believe that unlike 2008, this crisis will be more likely to redistribute wealth across banks than it is to create costs for the rest of us. Unlike 2008, when you could point to risk-seeking behavior on the part of banks as the prime reason for banking failures, this one was triggered by the search for high growth and a failure to adhere to first principles when it comes to duration mismatches. That said, I would expect the following changes in the banking structure:

1. Continued consolidation: Over the last few decades, the US banking business has consolidated, with the number of banks operating dropping 14,496 in 1984 to 4,844 in 2022. The 2023 bank failures will accelerate this consolidation, especially as small regional banks, with concentrated deposit bases and loan portfolios are assimilated into larger banks, with more diverse structure.
2. Bank profitability: For some, that consolidation is worrisome since it raises the specter of banks facing less competition and thus charging higher prices. I may be naive but I think that as banks consolidate, they will struggle to maintain profitability, and perhaps even see profits drop, as disruptors from fintech and elsewhere eat away at their most profitable segments. In short, the biggest banks may get bigger, but they may not get more profitable.
3. Accounting rule changes for banks: The fact that SVB's failure was triggered by a drop in value of the bank's investments in bonds and mortgage backed securities, and that this write down came as a shock to in investors, because SVB classified these securities as being held till maturity (and thus not requiring of mark-to-market) will inevitably draw the attention of accounting rule writers. While I don't foresee a requirement that every investment security be marked to market, a rule change that will create its own dangers, I expect the rules on when securities get marked to market to be tightened.
4. Regulatory changes: The 2023 crises have highlighted two aspects of bank behavior that are either ignored or sufficiently weighted into current regulatory rules on banks. The first is duration mismatches at banks, which clearly expose even banks that invest in default free securities, like SVB, to risk. The other is deposit stickiness, where old notions of when depositors panic and how quickly they react will have to be reassessed, given how quickly risk whispers about banks turned into deposit flight at First Republic and Signature Bank. I expect that there will be regulatory changes forthcoming that will try to incorporate both of these issues, but I remain unsure about the form that these changes will take.

I know I have said very little in this post about whether banks are good investments today, either collectively or subsets (large money center, regional etc.), and have focused mostly on what makes for good and bad banks. The reason is simple. Investing is not about judging the quality of businesses, but about buying companies at the right price, and that discussion requires a focus on what expectations markets are incorporating into stock prices. I will address the investing question in my next post, which I hope to turn to very soon! 

YouTube Video