Data Update 6 for 2025: From Macro to Micro - The Hurdle Rate Question!

    In the first five posts, I have looked at the macro numbers that drive global markets, from interest rates to risk premiums, but it is not my preferred habitat. I spend most of my time in the far less rarefied air of corporate finance and valuation, where businesses try to decide what projects to invest in, and investors attempt to estimate business value. A key tool in both endeavors is a hurdle rate – a rate of return that you determine as your required return for business and investment decisions. In this post, I will drill down to what it is that determines the hurdle rate for a business, bringing in what business it is in, how much debt it is burdened with and what geographies it operates in.

The Hurdle Rate - Intuition and Uses

    You don't need to complete a corporate finance or valuation class to encounter hurdle rates in practice, usually taking the form of costs of equity and capital, but taking a finance class both deepens the acquaintance and ruins it. It deepens the acquaintance because you encounter hurdle rates in almost every aspect of finance, and it ruins it, by making these hurdle rates all about equations and models. A few years ago, I wrote a paper for practitioners on the cost of capital, where I described the cost of capital as the Swiss Army knife of finance, because of its many uses. 

    In my corporate finance class, where I look at the first principles of finance that govern how you run a business, the cost of capital shows up in every aspect of corporate financial analysis:

• In business investing (capital budgeting and acquisition) decisions, it becomes a hurdle rate for investing, where you use it to decide whether and what to invest in, based on what you can earn on an investment, relative to the hurdle rate. In this role, the cost of capital is an opportunity cost, measuring returns you can earn on investments on equivalent risk.
  
  
  
• In business financing decisions, the cost of capital becomes an optimizing tool, where businesses look for a mix of debt and equity that reduces the cost of capital, and where matching up the debt (in terms of currency and maturity) to the assets reduces default risk and the cost of capital. In this context, the cost of capital become a measure of the cost of funding a business:
  
  
  
• In dividend decisions, i.e., the decisions of how much cash to return to owners and in what form (dividends or buybacks), the cost of capital is a divining rod. If the investments that a business is looking at earn less than the cost of capital, it is a trigger for returning more cash, and whether it should be in the form of dividends or buybacks is largely a function of what shareholders in that company prefer:
  

The end game in corporate finance is maximizing value, and in my valuation class, where I look at businesses from the outside (as a potential investor), the cost of capital reappears again as the risk-adjusted discount rate that you use estimate the intrinsic value of a business. 

Much of the confusion in applying cost of capital comes from not recognizing that it morphs, depending on where it is being used. An investor looking at a company, looking at valuing the company, may attach one cost of capital to value the company, but within a company, but within a company, it may start as a funding cost, as the company seeks capital to fund its business, but when looking at investment, it becomes an opportunity cost, reflecting the risk of the investment being considered.

The Hurdle Rate - Ingredients

    If the cost of capital is a driver of so much of what we do in corporate finance and valuation, it stands to reason that we should be clear about the ingredients that go into it. Using one of my favored structures for understanding financial decision making, a financial balance sheet, a cost of capital is composed of the cost of equity and the cost of debt, and I try to capture the essence of what we are trying to estimate with each one in the picture below:

To go from abstractions about equity risk and default risk to actual costs, you have to break down the costs of equity and debt into parts, and I try to do so, in the picture below, with the factors that you underlie each piece:

As you can see, most of the items in these calculations should be familiar, if you have read my first five data posts, since they are macro variables, having nothing to do with individual companies.  

• The first is, of course, the riskfree rate, a number that varies across time (as you saw in post on US treasury rates in data update 4) and across currencies (in my post on currencies in data update 5). 
• The second set of inputs are prices of risk, in both the equity and debt markets, with the former measured by equity risk premiums, and the latter by default spreads. In data update 2, I looked at equity risk premiums in the United States, and expanded that discussion to equity risk premiums in the rest of the world in data update 5). In data update 4, I looked at movements in corporate default spreads during 2024.

There are three company-specific numbers that enter the calculation, all of which contribute to costs of capital varying across companies;

• Relative Equity Risk, i.e., a measure of how risky a company's equity is, relative to the average company's equity. While much of the discussion of this measure gets mired in the capital asset pricing model, and the supposed adequacies and inadequacies of beta, I think that too much is made of it, and that the model is adaptable enough to allow for other measures of relative risk.
  
  I am not a purist on this measure, and while I use betas in my computations, I am open to using alternate measures of relative equity risk.
• Corporate Default Risk, i.e, a measure of how much default risk there is in a company, with higher default risk translating into higher default spreads. For a fairly large subset of firms, a bond rating may stand in as this measure, but even in its absence, you have no choice but to estimate default risk. Adding to the estimation challenge is the fact that as a company borrows more money, it will play out in the default risk (increasing it), with consequences for both the cost of equity and debt (increasing both of those as well).
• Operating geographies:  The equity risk premium for a company does not come from where it is  incorporated but from where it does business, both in terms of the production of its products and services and where it generates revenue. That said, the status quo in valuation in much of the world seems to be to base the equity risk premium entirely on the country of incorporation, and I vehemently disagree with that practice:
  
  Again, I am flexible in how operating risk exposure is measured, basing it entirely on revenues for consumer product and business service companies, entirely on production for natural resource companies and a mix of revenues and production for manufacturing companies.

As you can see, the elements that go into a cost of capital are dynamic and subjective, in the sense that there can be differences in how one goes about estimating them, but they cannot be figments of your imagination.

The Hurdle Rate - Estimation in 2025

    With that long lead in, I will lay out the estimation choices I used to estimate the costs of equity, debt and capital for the close to 48,000 firms in my sample. In making these choices, I operated under the obvious constraint of the raw data that I had on individual companies and the ease with which I could convert that data into cost of capital inputs. 

1. Riskfree rate: To allow for comparisons and consolidation across companies that operate in different currencies, I chose to estimate the costs of capital for all companies in US dollars, with the US ten-year treasury rate on January 1, 2025, as the riskfree rate.
2. Equity Risk Premium: Much as I would have liked to compute the equity risk premium for every company, based upon its geographic operating exposure, the raw data did not lend itself easily to the computation. Consequently, I have used the equity risk premium of the country in which a company is headquartered to compute the equity risk premium for it.
3. Relative Equity Risk: I stay with beta, notwithstanding the criticism of its effectiveness for two reasons. First, I use industry average betas, adjusted for leverage, rather than the company regression beta, because because the averages (I title them bottom up betas) are significantly better at explaining differences in returns across stocks. Second, and given my choice of industry average betas, none of the other relative risk measures come close, in terms of predictive ability. For individual companies, I do use the beta of their primary business as the beta of the company, because the raw data that I have does not allow for a breakdown into businesses. 
4. Corporate default risk: For the subset of the sample of companies with bond ratings, I use the S&P bond rating for the company to estimate the cost of debt. For the remaining companies, I use interest coverage ratios as a first measure to estimate synthetic ratings, and standard deviation in stock prices as back-up measure.
5. Debt mix: I used the market capitalization to measure the market value of equity, and stayed with total debt (including lease debt) to estimate debt to capital and debt to equity ratios

The picture below summarizes my choices:

There are clearly approximations that I used in computing these global costs of capital that I would not use if I were computing a cost of capital for valuing an individual company, but this approach yields values that can yield valuable insights, especially when aggregated and averaged across groups.

a. Sectors and Industries

    The risks of operating a business will vary  widely across different sectors, and I will start by looking at the resulting differences in cost of capital, across sectors, for global companies:

There are few surprises here, with technology companies facing the highest costs of capital and financials the lowest, with the former pushed up by high operating risk and a resulting reliance on equity for capital, and the latter holding on because of regulatory protection.

    Broken down into industries, and ranking industries from highest to lowest costs of capital, here is the list that emerges:

Download industry costs of capital

The numbers in these tables may be what you would expect to see, but there are a couple of powerful lessons in there that businesses ignore at their own peril. The first is that even a casual perusal of differences in costs of capital across industries indicates that they are highest in businesses with high growth potential and lowest in mature or declining businesses, bringing home again the linkage between danger and opportunity. The second is that multi-business companies should understand that the cost of capital will vary across businesses, and using one corporate cost of capital for all of them is a recipe for cross subsidization and value destruction.

b. Small versus Larger firms

    In my third data update for this year, I took a brief look at the small cap premium, i.e, the premium that small cap stocks have historically earned over large cap stocks of equivalent risk, and commented on its disappearance over the last four decades. I heard from a few small cap investors, who argued that small cap stocks are riskier than large cap stocks, and should earn higher returns to compensate for that risk. Perhaps, but that has no bearing on whether there is a small cap premium, since the premium is a return earned over and above what you would expect to earn given risk, but I remained curious as to whether the conventional wisdom that small cap companies face higher hurdle rates is true. To answer this question, I examine the relationship between risk and market cap, breaking companies down into market cap deciles at the start of 2025, and estimating the cost of capital for companies within each decile:

The results are mixed. Looking at the median costs of capital, there is no detectable pattern in the cost of capital, and the companies in the bottom decile have a lower median cost of capital (8.88%) than the median company in the sample (9.06%). That said, the safest companies in  largest market cap decile have lower costs of capital than the safest companies in the smaller market capitalizations. As a generalization, if small companies are at a disadvantage when they compete against larger companies, that disadvantage is more likely to manifest in difficulties growing and a higher operating cost structure, not in a higher hurdle rate.

c. Global Distribution

    In the final part of this analysis, I looked at the costs of capital of all publicly traded firms and played some Moneyball, looking at the distribution of costs of capital across all firms. In the graph below,I present the histogram of cost of capital, in US dollar terms, of all global companies at the start of 2025, with a breakdown of costs of capital, by region, below:

I find this table to be one of the most useful pieces of data that I possess and I use it in almost every aspect of corporate finance and valuation:

1. Cost of capital calculation: The full cost of capital calculation is not complex, but it does require inputs about operating risk, leverage and default risk that can be hard to estimate or assess for young companies or companies with little history (operating and market). For those companies, I often use the distribution to estimate the cost of capital to use in valuing the company. Thus, when I valued Uber in June 2014, I used the cost of capital (12%) at the 90th percentile of US companies, in 2014, as Uber's cost of capital. Not only did that remove a time consuming task from my to-do list, but it also allowed me to focus on the much more important questions of  revenue growth and margins for a young company. Drawing on my fifth data update, where I talk about differences across currencies, this table can be easily modified into the currency of your choice, by adding differential inflation. Thus, if you are valuing an Indian IPO, in rupees, and you believe it is risky, at the start of 2025, adding an extra 2% (for the inflation differential between rupees and dollars in 2025) to the ninth decile of Indian costs of capital (12.08% in US dollars) will give you a 14.08% Indian rupee cost of capital.
2. Fantasy hurdle rates: In my experience, many  investors and companies make up hurdle rates, the former to value companies and the latter to use in investment analysis. These hurdle rates are either hopeful thinking on the part of investors who want to make that return or reflect inertia, where they were set in stone decades ago and have never been revisited. In the context of checking to see whether a valuation passes the 3P test (Is it possible? Is it plausible? Is it probable?), I do check the cost of capital used in the valuation. A valuation in January 2025, in US dollars, that uses a 15% cost of capital for a publicly traded company that is mature is fantasy (since it is in well in excess of the 90th percentile), and the rest of the valuation becomes moot. 
3. Time-varying hurdle rates: When valuing companies, I believe in maintaining consistency, and one of the places I would expect it to show up is in hurdle rates that change over time, as the company's story changes. Thus, if you are valuing a money-losing and high growth company, you would expect its cost of capital to be high, at the start of the valuation, but as you build in expectations of lower growth and profitability in future years, I would expect the hurdle rate to decrease (from close to the ninth decile in the table above towards the median).

It is worth emphasizing that since my riskfree rate is always the current rate, and my equity risk premiums are implied, i.e., they are backed out from how stocks are priced, my estimates of costs of capital represent market prices for risk, not theoretical models. Thus, if looking at the table, you decide that a number (median for your region, 90th percentile in US) look too low or too high, your issues are with the market, not with me (or my assumptions).

Takeaways

    I am sorry that this post has gone on as long as it has, but to end, there are four takeaways from looking at the data:

1. Corporate hurdle rate: The notion that there is a corporate hurdle rate that can be used to assess investments across the company is a myth, and one with dangerous consequences. It plays out in all divisions in a multi-business company using the same (corporate) cost of capital and in acquisitions, where the acquiring firm's cost of capital is used to value the target firm. The consequences are predictable and damaging, since with this practice, safe businesses will subsidize risky businesses, and over time, making the company riskier and worse off over time.
2. Reality check on hurdle rates: All too often, I have heard CFOs of companies, when confronted with a cost of capital calculated using market risk parameters and the company's risk profile, say that it looks too low, especially in the decade of low interest rates, or sometimes, too high, especially if they operate in an risky, high-interest rate environment. As I noted in the last section, making up hurdle rates (higher or lower than the market-conscious number) is almost never a good idea, since it violates the principle that you have live and operate in the world/market you are in, not the one you wished you were in.
3. Hurdle rates are dynamic: In both corporate and investment settings, there is this almost desperate desire for stability in hurdle rates. I understand the pull of stability, since it is easier to run a business when hurdle rates are not volatile, but again, the market acts as a reality check. In a world of volatile interest rates and risk premia, using a cost of capital that is a constant is a sign of denial.
4. Hurdle rates are not where business/valuation battles are won or lost: It is true that costs of capital are the D in a DCF, but they are not and should never be what makes or breaks a valuation. In my four decades of valuation, I have been badly mistaken many times, and the culprit almost always has been an error on forecasting growth, profitability or reinvestment (all of which lead into the cash flows), not the discount rate. In the same vein, I cannot think of a single great company that got to greatness because of its skill in finessing its cost of capital, and I know of plenty that are worth trillions of dollars, in spite of never having actively thought about how to optimize their costs of capital. It follows that if  you are spending the bulk of your time in a capital budgeting or a valuation, estimating discount rates and debating risk premiums or betas, you have lost the script. If you are valuing a mature US company at the start of 2025, and you are in a hurry (and who isn't?), you would be well served using a cost of capital of 8.35% (the median for US companies at the start of 2025) and spending your time assessing its growth and profit prospects, and coming back to tweak the cost of capital at the end, if you have the time.

YouTube Video

Data Updates for 2025

1. Data Update 1 for 2025: The Draw (and Danger) of Data!
2. Data Update 2 for 2025: The Party continued for US Equities
3. Data Update 3 for 2025: The times they are a'changin'!
4. Data Update 4 for 2025: Interest Rates, Inflation and Central Banks!
5. Data Update 5 for 2025: It's a small world, after all!
6. Data Update 6 for 2025: From Macro to Micro - The Hurdle Rate Question!

Data Links

1. Cost of capital, by industry grouping: US, Global, Emerging Markets, Japan, Europe, India, China)
2. Cost of capital distribution, by industry

Paper links

1. The Cost of Capital: The Swiss Army Knife of Finance

Data Update 5 for 2025: It's a small world, after all!

If the title of this post sounds familiar, it is because is one of Disney’s most iconic rides, one that I have taken hundreds of times, first with my own children and more recently, with my grandchildren. It is a mainstay of every Disney theme park, from the original Disneyland in Anaheim to the newer theme parks in Paris, Hong Kong and Shanghai. For those of who have never been on it, it is the favored ride for anyone who is younger than five in your group, since you spend ten minutes in a boat going through the world as Disney would like you to see it, full of peace, happiness, and goodwill. In this post, I will expand my analysis of data in 2024, which has a been mostly US-centric in the first four of my posts, and use that data to take you on my version of the Disney ride, but on this trip, I have no choice but to face the world as is, with all of the chaos it includes, with tariffs and trade wars looming. 

Returns in 2024

    Clearly, the most obvious place to start this post is with market performance, and in the table below, I report the percentage change in index level, for a subset of indices, in 2024:

The best performing index in 2024, at least for the subset of indices that I looked at, was the Merval, up more than 170% in 2024, and that European indices lagged the US in 2024. The Indian and Chinese markets cooled off in 2024, posting single digit gains in price appreciation. 

   There are three problems with comparing returns in indices. First, they are indices and reflect a subset of stocks in each market, with different criteria determining how each index is constructed, and varying numbers of constituents. Second, they are in local currencies, and in nominal terms. Thus, the 172.52% return in the Merval becomes less impressive when inflation in Argentina is taken into account. It is for this reason that I chose to compute returns differently, using the following constructs:

1. I included all publicly traded stocks in each market, or at least those with a market capitalization available for them.
2. I converted all of the market capitalizations into US dollars, just to make them comparable.
3. I aggregated the market capitalizations of all stocks at the end of 2023 and the end of 2024, and computed the percentage change.

The results, broken down broadly by geography are in the table below:

As you can see, the aggregate market cap globally was up 12.17%, but much of that was the result of a strong US equity market. Continuing a trend that has stretched over the last two decades, investors who tried to globally diversify in 2024 underperformed investors who stayed invested only in the United States. 

    I do have the percentage changes in market cap, by country, but you should take those results with a grain of salt, since there are countries with just a handful of listings, where the returns are distorted. Looking at countries with at least ten company listings, I have a list of the ten best and worst performing countries in 2024:

Argentina's returns in US dollar terms is still high enough to put it on top of the list of best-performing countries in the world in 2024 and Brazil is at the top of the list of worst performing countries, at least in US dollar terms.

The Currency Effect

    As you can see comparing the local index and dollar returns, the two diverge in some parts of the world, and the reason for the divergence is movements in exchange rates. To cast light on this divergence, I looked at the US dollar's movements against other currencies, using three variants of US dollar indices against emerging market currencies, developed market currencies and broadly against all currencies:

FRED

The dollar strengthened during 2024, more (10.31%) against emerging market currencies than against developed market currencies (7.66%), and it was up broadly (9.03%).

    I am no expert on exchange rates, but learning to deal with different currencies in valuation is a prerequisite to valuing companies. Since I value companies in local currencies, I am faced with the task of estimating risk free rates in dozens of currencies, and the difficulty you face in estimating these rates can vary widely (and be close to impossible in some) across currencies. In general, you can break down risk free estimation, in different currencies, in three groupings, from easiest to most difficult:

My process for estimating riskfree rates in a currency starts with a government issuing a long term bond in that currency, and if the government in question has no default risk, it stops there. Thus, the current market interest rate on a long term Swiss government bond, in Swiss Francs, is the risfree rate in that currency. The process gets messier, when there is a long-term, local currency bond that is traded, but the government issuing the bond has default risk. In that case, the default spread on the bond will have to be netted out to get to a riskfree rate in the currency.  There are two key estimation questions that are embedded in this approach to estimating riskfree rates. The first is the assessment of whether there is default risk in a government, and I use a simplistic (and flawed) approach, letting the local currency sovereign rating for the government stand in as the measure; I assume that AAA rated government bonds are default-free, and that any rating below is a indication of default risk. The second is the estimation of the default spread, and in my simplistic approach, I use one of two approaches - a default spread based upon the sovereign rating or a sovereign credit default swap spread. At the start of 2025, there were just about three dozen currencies, where I was able to find local-currency government bonds, and I estimated the riskfree rates in these currencies;

Download data

At the risk of stating the obvious (and repeating what I have said in earlier posts), there is no such thing as a global riskfree rate, since riskfree rates go with currencies, and riskfree rates vary across currencies, with all or most of the difference attributable to differences in expected inflation. High inflation currencies will have high riskfree rates, low inflation currencies low riskfree rates and deflationary currencies can negative riskfree rates.

    It is the recognition that differences in riskfree rates are primarily due to differences in expected inflation that gives us an opening to estimate riskfree rates in currencies without a government bond rate, or even to run a sanity check on the riskfree rates that you get from government bonds. If you start with a riskfree rate in a currency where you can estimate it (say US dollars, Swiss Francs or Euros), all you need to estimate a riskfree rate in another currency is the differential inflation between the two currencies. Thus, if the US treasury bond rate (4.5%) is the riskfree rate in US dollars, and the expected inflation rates in US dollars and Brazilian reals are 2.5% and 7.5% respectively, the riskier rate in Brazilian reals:

Riskfree rate in $R = (1+ US 10-year T.Bond Rate) * (1 + Expected inflation rate in $R)/ (1+ Expected inflation rate in US $) - 1 = 1.045 *(1.075/1.025) -1 = 9.60%

In approximate terms, this can be written as

Riskfree rate in $R = US 10-year T.Bond Rate + (Expected inflation rate in $R) - Expected inflation rate in US $) - 1 = 4.5% - (7.5% - 2.5%) = 9.50%

While obtaining an expected inflation rate for the US dollar is easy (you can use the difference between the ten-year US treasury bond rate and the ten-year US TIPs rate), it can be more difficult to obtain this number in Egyptian pounds or in Zimbabwean dollars, but you can get estimates from the IMF or the World Bank. 

The Risk Effect

    There are emerging markets that have delivered higher returns than developed markets, but in keeping with a core truth in investing and business, these higher returns often go hand-in-hand with higher risk. The logical step in looking across countries is measuring risk in countries, and bringing that risk into your analysis, by incorporating that risk by demanding higher expected returns in riskier countries.

    That process of risk analysis and estimating risk premiums starts by understanding why some countries are riskier than others. The answers, to you, may seem obvious, but I find it useful to organize the obvious into buckets for analysis. I will use a picture in posts on country risk before to capture the multitude of factors that go into making some countries riskier than others:

To get from these abstractions to country risk measures, I make a lot of compromises, putting pragmatism over purity. While I take a deeper look at the different components of country risk in my annual updates on country risk (with the most recent one from 2024), I will cut to the chase and focus explicitly on my approach to estimating equity risk premiums, using my 2025 data update to illustrate:

With this approach, I estimated equity risk premiums, by country, and organized by region, here is what the world looked like, at the start of 2025:

Download equity risk premiums by country

Note that I attach the implied equity risk premium for the S&P 500 of 4.33% (see my data update 3 from a couple of weeks ago) to all Aaa rated countries (Australia, Canada, Germany etc.) and an augmented premium for countries that do not have Aaa ratings, with the additional country risk premium determined by local currency sovereign ratings. 

    I am aware of all of the possible flaws in this approach. First, treating the US as default-free is questionable, now that it has threatened default multiple times in the last decade and has lost its Aaa rating with every ratings agency, other than Moody's. That is an easily fixable problem, though, since if you decide to use S&P's AA+ rating for the US, all it would require is that you net out the default spread of 0.40% (for a AA+ rating at the start of 2025) from the US ERP to get a mature market premium of 3.93% (4.33% minus 0.40%). Second, ratings agencies are not always the best assessors of default risk, especially when there are dramatic changes in a country, or when they are biased (towards or against a region). That too has a fix, at least for the roughly 80 countries where there are trade sovereign CDS spreads, and those sovereign CDS spreads can be used instead of the ratings-based spreads for those countries.

The Pricing Effect

   As an investor, the discussions about past returns and risk may miss the key question in investing, which is pricing. At the right price, you should be willing to buy stocks even in the riskiest countries, and especially so after turbulent (down) years. At the wrong price, even the safest market with great historical returns are bad investments. To assess pricing in markets, you have to scale the market cap to operating metrics, i.e., estimate a multiple, and while easy enough to do, there are some simple rules to follow in pricing. 

    The first is recognizing that every multiple has a market estimate of value in the numerator, capturing either just equity value (market cap of equity), total firm value (market cap of equity + total debt) or operating asset (enterprise) value (market cap of equity + total debt - cash):

Depending on the scalar (revenues, earnings, book value or cash flow), you can compute a variety of multiples, and if you add on the choices on timing for the scaling variables (trailing, current, forward), the choices multiply. To the question of which multiple is best, a much debated topic among analysts, my answer is ambivalent, since you can use any of them in pricing, as long as you ask the right follow-up questions. 

    To compare how stocks are priced globally, I will use three of these multiples. The first is the price earnings ratio, partly because in spite of all of its faults, it remains the most widely used pricing metric in the world. The second is the polar opposite on the pricing spectrum, which is the enterprise value to sales multiple, where rather than focus on just equity value, I look at operating asset value, and scale it to the broadest of operating metrics, which is revenue. While it takes a lot to get from revenues to earnings, the advantage of using revenues is that it is number least susceptible to accounting gaming, and also the one where you are least likely to lose companies from your sample. (Thousands and thousands of companies in my sample have negative net income, making trailing PE not meaningful, but very few (usually financial service firms) have missing revenues). The third pricing metric I look at is the enterprise value to EBITDA, a multiple that has gone from being lightly used four decades ago to a banking punchline today, where EBITDA represents a rough measure of operating cash flow). With each of these multiples, I make two estimation choices:

1. I stay with trailing values for net income, revenues and EBITDA, because too many of the firms in my 48,000 firm sample have no analysts following them, and hence no forward numbers.
2. I compute two values for each country (region), an aggregated version and the median value. While the latter is simple, i.e., it is the median number across all companies in a country or region, the former is calculated across all companies, by aggregating the values across companies. Thus, the aggregated PE ratio for the United States is 20.51, and it computed by adding up the market capitalizations of all traded US stocks and dividing by the sum of the net income earned by all traded firms, including money losers. Think of it a weighted-average PE, with no sampling bias.

With these rules in place, here is what the pricing metrics looked like, by region, at the start of 2025:

The perils of investing based just upon pricing ratios should be visible from this table. Two of the cheapest regions of the world to invest in are Latin America and Eastern Europe, but both carry significant risk with them, and the third, Japan, has an aging population and is a low-growth market. The most expensive market in the world is India, and no amount of handwaving about the India story can justify paying 31 times earnings, 3 times revenue and 20 times EBITDA, in the aggregate, for Indian companies. The US and China also fall into the expensive category, trading at much higher levels than the rest of the world, on all three pricing metrics.

    Within each of these regions, there are differences across countries, with some priced more richly than others. In the table below, I look at the ten countries, with at least 5 companies listed on their exchanges, that trade at the lowest median trailing PE ratios, and the ten countries that are more expensive using that same metric:

Many of the markets are in the world that trade at the lowest multiples of trailing earnings are in Africa. With Latin America, it is a split decisions, where you have two countries (Colombia and Brazil) on the lowest PE list and one (Argentina) on the highest PE list. In some of the countries, there is a divergence between the aggregated version and the trailing PE, with the aggregated PE higher (lower) than the median value, reflecting larger companies that trade at lower (higher) PE ratios than the rest of the market.

    Replacing market cap with enterprise value, and net income with revenues, gives you a pricing multiple that lies at the other end of the spectrum, and ranking countries again, based on median EV to sales multiples, here is the list of the ten most expensive and cheapest markets:

On an enterprise value to sales basis, you see a couple of Asian countries (Japan and South Korea) make the ten lowest list, but the preponderance of Middle Eastern countries on ten highest lists may just be a reflection of quirks in sample composition (more financial service firms, which have no revenues, in the sample).

The Year to come

    This week has been a rocky one for global equities, and the trigger for the chaos has come from the United States. The announcements, from the Trump administration, of the intent to impose 25% tariffs on Canada and Mexico may have been delayed, and perhaps may not even come into effect, but it seems, at least to me, a signal that globalization, unstoppable for much of the last four decades, has crested, and that nationalism, in politics and economics, is reemerging. 

    As macroeconomists are quick to point out, using the Great Depression and Smoot-Hawley's tariffs in the 1930 to illustrate, tariffs are generally not conducive to global economic health, but it is time that they took some responsibility for the backlash against free global trade and commerce. After all, the notion that globalization was good for everyone was sold shamelessly, even though globalization created winners (cities, financial service firms) and losers (urban areas, developed market manufacturing) , and much of what we have seen transpired over the last decade (from Brexit to Trump) can be viewed as part of the backlash. In spite of the purse clutching at the mention of tariffs, they have been part of global trade as long as there has been trade, and they did not go away after the experiences with the depression. I agree that the end game, if tariffs and trade wars become commonplace, will be a less vibrant global economy, but as with any major macroeconomic shocky, there will be winners and losers. 

    There is, I am sure, a sense of schadenfreude among many in emerging markets, as they watch developed markets start to exhibit the behavior (unpredictable government policy, subservient central banks, breaking of legal and political norms) that emerging markets were critiqued for decades ago, but the truth is that the line between developed and emerging markets has become a hazy one. After the fall of the Iron Curtain, George H.W. Bush (the senior) declared a "new world order", a proclamation turned out to be premature, since the old world order quickly reasserted itself. The political and economic developments of the last decade may signal the arrival of a new world order, though no one in quite sure whether it will be better or worse than the old one. 

YouTube Video

Data Updates for 2025

1. Data Update 1 for 2025: The Draw (and Danger) of Data!
2. Data Update 2 for 2025: The Party continued for US Equities
3. Data Update 3 for 2025: The times they are a'changin'!
4. Data Update 4 for 2025: Interest Rates, Inflation and Central Banks!
5. Data Update 5 for 2025: It's a small world, after all!

Data Links

1. Riskfree rates, by currency, in January 2025
2. Equity risk premiums, by country, in January 2025
3. Pricing ratios, by country, in January 2025

DeepSeek crashes the AI Party: Story Break, Change or Shift?

    I am going to start this post with a confession that my knowledge of the architecture and mechanics of AI are pedestrian and that there will be things that I don't get right in this post. That said, DeepSeek's abrupt entry into the AI conversation has the potential to change the AI narrative, and as it does, it may also change the storylines for the many companies that have spent the last two years benefiting from the AI hype. I first posted about AI in the context of valuing Nvidia, in June 2023, when there was still uncertainty about whether AI had legs. A little over a year later, in September 2024, that question about AI seemed to have been answered in the affirmative, for most investors, and I posted again after Nvidia had a disappointing earnings report, arguing that it reflected a healthy scaling down of expectations. As talk of AI disrupting jobs and careers also picked up, I also posted a piece on the threat that AI poses for all of us, with its capacity to do our jobs, at low or no cost, and what I saw as the edges I could use to keep my bot at bay. For those of you who have been tracking the market, the AI segment in the market has held its own since September, but even before the last weekend, there were signs that investors were sobering up on not only how big the payoff to AI would be, but how long they would have to wait to get there. 

The AI story, before DeepSeek

    The AI story has been building for a while, reflecting the convergence of two forces in technology - more computing power, often in smaller and smaller packages, and the accumulation of data, on technology platforms and elsewhere. That said, the AI story broke out to the public on November 30, 2022, when OpenAI launched ChatGPT, and it made its presence felt in homes, schools and businesses almost instantaneously. It is that wide presence in our daily lives that laid the foundations for the AI story, where evangelists sold us on the notion that AI solutions would make our lives easier and take away the portions of our work that we found most burdensome, and that the businesses that provided these solutions would be worth trillions of dollars.

    As the number of potential applications of AI proliferated, thus increasing the market for AI products and services, another part of the story was also being put into play. AI was framed as being made possible by the marriage of incredibly powerful computers and deep troves of data, effectively setting the stage for the winners, losers, and wannabes in the story. The first set of companies were perceived as benefiting from building the AI architecture, with the advance spending on this architecture coming from the companies that hoped to be players in the AI product and service markets:

1. Computing Power: In the AI story that was told, the computers that were needed were so powerful that they needed customized chips, more powerful and compact than any made before, and one company (Nvidia), by virtue of its early start and superior chip design capabilities, stood well above the rest. Not only did Nvidia have an 80% market share of the AI chip market, as assessed in 2024, the lead and first-mover advantage that the company possessed would give it a dominant market share, in the much larger AI chip market of the future. Along the way, the the AI story picked up supercomputing companies, as passengers, again on the belief that Ai systems would find a use for them.
2. Power: In the AI story, the coupling of powerful computing and immense data happens in data centers that are power hogs, requiring immense amounts of energy to keep going. Not surprisingly, a whole host of power companies have stepped into the breach, with some increasing capacity entirely to service these data centers. Some of them were new entrants (like Constellation Energy), whereas others were more traditional power companies (Siemens Energy) who saw an opening for growth and profitability in the AI space. 
3. Data: A third beneficiary from the architecture part of the AI story were the cloud businesses, where the big data, collected for the AI systems would get stored. The big tech companies with cloud arms, particularly Microsoft (Azure) and Amazon (AWS) have benefited from that demand, as have other cloud businesses.

Since the companies involved in building the AI infrastructure are the ones that are most tangibly (and immediately) benefiting from the AI boom, they are also the companies that have seen the biggest boost in market cap, as the AI story heated up. In the graph, I have picked on a subset of high-profile companies that were part of the AI market euphoria and looked at the consequent increase in their market capitalizations:

Using the ChatGPT introduction on November 30, 2022, as the starting point for the AI buzz, in public consciousness and markets, the returns in 2023 and 2024 are a composite (albeit a rough) measure of the benefits that AI has generated for these companies. Note that the biggest percentage winner, at least in this group was Palantir, up 1285% in the last two years, but the biggest winner in absolute terms was Nvidia, which gained almost $ 3 trillion in value in 2023 and 2024.

    The investments in that AI architecture were being made, with the expectation that companies that invested in the architecture would be able to eventually profit from developing and selling AI products and services. Since the AI storyline required immense upfront investing in computing power and access to big data, the biggest investors in AI architecture were big tech companies, with Microsoft and Meta being the largest customers for Nvidia chips in 2024. In the table below, I look at the Mag Seven, not inclusive of Nvidia, and examine the returns that they have made in 2023 and 2024:

As you can see, the Mag Seven carried the market in the two years, each adding a trillion (or close, in the case of Tesla) dollars in value in the last two years, with some portion of that value attributable to the AI story. With requirements for large investment up front acting as entry barriers, the expectation was these big tech companies would eventually not only be able to develop AI products and services that their customers would want, but charge premium prices (and earn higher margins).

    In the picture below, I have tried to capture the essence of AI story, with the potential winners and losers at each stage:

There are parts to this story where there is much to be proved, especially on the AI product and service part, and while investors can be accused of becoming excessively exuberant about the story, it is a plausible one. In fact, my most recent (in September 2024) valuation of Nvidia bought into core elements of the story, though I still found it overvalued:

Nvidia valuation in September 2024 (Pre DeepSeek)

Note that the big AI story plays out in these inputs in multiple places:

1. AI chip market: My September 2024 estimate for the size of the AI chip market was $500 billion, which in turn was justifiable only because the AI product and service market was expected to huge ($3 trillion and beyond).
2. Nvidia market share: In my valuation, I assumed that Nvidia's lead in the AI chip business would give the company a head start, as the business grew, and to the extent that demand is sticky (i.e., once companies start build data centers with Nvidia chips, it would be difficult for them to switch to a competitor), Nvidia would maintain a dominant market share (60%) of the expanded AI chip market.
3. Nvidia margins: Nvidia has had immense pricing power, posting nosebleed-level gross and operating margins, while TSMC (its chip maker) has generated only a fraction of the benefits, and its biggest customers (the big tech companies) have been willing to pay premium prices to get a head start in building their AI architecture. Over time, I assumed that Nvidia would see its margins drop, but even with the drop, their target margin (60%) would resemble those of very successful, software companies, not chip making companies.

My concern in September 2024, and in fact for the bulk of the last two years, was not that I had doubts about the core AI story, but that investors were overpaying for the story. That is partly why, I have shed portions of my holdings in Nvidia, selling half my holdings in the summer of 2023 and another quarter in the summer of 2024.

The AI Story, after DeepSeek

    I teach valuation, and have done so for close to forty years. One reason I enjoy the class is that you are never quite done with a valuation, because life keeps throwing surprises at you. The first session of my undergraduate valuation class was last Wednesday (January 22), and during the course of the class, I talked about how a good valuation connects narrative to numbers, and followed up by noting that even the most well thought through narratives will change over time. I am not sure how much of that message got through to my studentls, but the message was delivered much more effectively by DeepSeek's entry into the AI story over the weekend, and the market shakeup that followed when markets opened on Monday (January 27).

A DeepSeek Primer

    The DeepSeek story is still being told, and there is much we do not know. For the moment, though, here is what we know. In 2010, Liang Wenfeng, a software engineer, founded DeepSeek as a hedge fund in China, with the intent of using artificial intelligence to make money. Unable to get traction in that endeavor, and facing government hostility on speculative trading, he pivoted in 2023 into AI, putting together a team to create a Chinese competitor to OpenAI. Since the intent was to come up with a product that could be sold at bargain prices, DeepSeek did what disruptors have always done, which is look for an alternate path to the same destination (providing AI products that work). Rather than invest in expensive infrastructure (supercomputers and data centers), DeepSeek used much cheaper, less powerful chips, and instead of using immense amounts of data, created an AI prototype that could work with less data, using rule-based logic to fill in the gap. While there has been chatter about DeepSeek for weeks, it became publicly accessible at the end of last week (ending January 24), and within hours, was drawing rave reviews from people well versed in tech, as it matched beat ChatGPT at many tasks, and even performed better on scientific and math queries. 

    There are parts of this story that are clearly for public consumption, more side stories than main story,, and it is best to get them out of the way, before looking at the DeepSeek effect.

1. Cost of development: The notion that DeepSeek was developed for just a few million dollars is fantasy, and while there may have been a portion of the development that cost little, the total was probably in the hundreds of millions of dollars and required a lot more resources (including perhaps even Nvidia chips) than the developers are letting on. No matter what the true cost of development is finally revealed to be, it will be a fraction of the money spent by the existing players in building their systems.
2. Performance tests: The tests of DeepSeek versus OpenAI (or Claude and Gemini) suggests that DeepSeek not only holds it own against the establishment, but even outperforms them on some tasks. That is impressive, but the leap that some are making to concede the entire AI product and service market to DeepSeek is unwarranted. There are clearly aspects of the AI products and service business, where the DeepSeek approach (of using less powerful computing and data) will be good enough, but there will be other aspects of the AI business, where the old paradigm of super computing power and vast data will still hold.
3. A Chinese company: The fact that DeepSeek was developed in China throws a political twist into the story that will undoubtedly play a role in how it develops, but the genie is out of the bottle, even if other governments try to stop its adoption. Adding to the noise is the decision by the company to make DeepSeek open-source, effectively allowing others to adapt and build their own versions.
4. Fair or foul: Finally, there has been some news on the legal front, where OpenAI has argued that DeepSeek unlawfully used data that was generated by OpenAI in building their offering, and while part of that lawsuit may just be showboating, it is possible that portions of the story are true and that legal consequences will follow.

While we can debate the what's and why's in this story, the market reaction this week to the story has been swift and decisive. I graph the performance of the five AI stocks highlighted in the earlier section, throwing in the Meta and Microsoft for good measure, on a daily basis in 2025.

As you can see in this chart, Nvidia Broadcom, Constellation and Vistra have had terrible weeks, losing more than 10% in the last week, but just for perspective, also note that Constellation and Vistra are still up strongly for the year. Meta and Microsoft were unaffected, and so was Palantir, Clearly, the DeepSeek story is playing out differently for different companies in the AI space, but its overall market impact has been substantial, and for the most part, negative.

    What is it that makes the DeepSeek story so compelling? First, is the technological aspect of coming up with a product, with far less in resources that the establishment, and I have nothing but admiration for the DeepSeek creators, but the part of the story that stands out is that the they chose not to go with the prevailing narrative (the one where Nvidia chips and huge data bases are a necessity) and instead asked the question of what the end products and services would look like, and whether there was an easier, quicker and cheaper way of getting there. In hindsight, there are probably others who are looking at DeepSeek and wondering why they did not choose the same path, and the answer is that it takes courage to go against the conventional wisdom, especially when, as AI did over the last two years, it sweeps everyone (from tech titans to individual investors) along with its force.

    The truth is that even if DeepSeek is stopped through legal or government action or fails to deliver on its promises, what its entry has done to the AI story cannot be undone, since it has broken the prevailing narrative. I would not be surprised if there are a dozen other start-ups, right now, using the DeepSeek playbook to come up with their own lower-cost competitors to prevailing players. Put simply, the AI story's weakest links have been exposed, and if this were the tale about the Emperor's new clothes, the AI emperor is, if not naked, is having a wardrobe malfunction, for all to see.

The Story Effect

    In this first week, as is to be expected, the response has been anything but reasoned. If you are a voracious reader of financial news (I am not), you have probably seen dozens of “thought pieces” from both technology and market experts claiming to foretell the future, and even among the few that I have read, the views range the spectrum on how DeepSeek changes the AI story. 

    In my writings on narrative and numbers, where I talk about how every valuation tells a story, I also talk about how stories are dynamic, with a story break representing radical change (where a great story can crash and burn or a small story can break out to become a big one), a story change can be a significant narrative alteration (where a story adds or loses a dimension with big value effects) or a story shift (where the core story remains unchanged, but the parameters can change). Using the pre-DeepSeek story as a starting point, you can classify the narratives on what is coming on the story break/story change/story shift continuum:

With all the caveats, including the fact that I am an AI novice, with a deeper understanding of potato chips than computer chips, and that it is early in the game, I am going to take a stand on where in this continuum I see the DeepSeek effect falling. I believe that DeepSeek does change the AI story, by creating two pathways to the AI product and service endgame. On one path that will lead to what I will term the “low intensity” AI market, it has opened the door to lower cost alternatives, in terms of investments in computing power and data, and competitors will flock in. That said, there will remain a segment of the AI market, where the old story will prevail, and the path of massive investments in computer chips and data centers leading to premium AI products and services will be the one that has to be taken.

    Note that the entry characteristics for the two paths will also determine the profitability and payoffs from their respective AI product and service markets (that will eventually exist). The “low entry cost” pathway is more likely to lead to commoditization, with lots of competitors and low pricing power, whereas the “high entry cost” path with its requirements for large upfront investment and access to data will create a more restrictive market, with higher priced and more profitable AI products and services. This story leaves me with a judgment call to make about the relative sizes of the markets for the two pathways. I am generalizing, but much of what consumers have seen so far as AI offerings fall into the low cost pathway and I would not be surprised, if that remains true for the most part. The DeepSeek entry has now made it more likely that you and I (as consumers) will see more AI products and services offered to us, at low cost or even for free. There is another segment of the AI products and services market, though, with businesses (or governments) as customers, where significant investments made and refinements will lead to AI products and services, with much higher price points. In this market, I would not be surprised to see networking benefits manifest, where the largest players acquire advantages, leading to winner-take-all markets. 

    In telling this story, I understand that not only am I going to be wrong, perhaps decisively, but also that it could unravel in record time. I make this leap, not out of arrogance or a misplaced desire to change how you think, but because I own a slice of Nvidia (one quarter of the holding that I had two years ago, but still large enough to make a difference in my portfolio), and I cannot value the company without an AI story in place. That said, the feedback loop remains open, and I will listen not only to alternate opinions but also follow real world developments, in the interests of telling a better story.

The Value Effect

    Now that my AI story is in the open, I will use it to revisit my valuation of Nvidia, and incorporate my new AI story in that valuation. Even without working through the numbers, it is very difficult to see a scenario where the entry of DeepSeek makes Nvidia a more valuable company, with the biggest change being in the expected size of the AI chip market:

	In September 2024 (pre DeepSeek)	In January 2025 (post DeepSeek)
AI chip market size in 2035	$500 billion	$300 billion
Nvidia's market share	60%	60%
Nvidia's operating margin	60%	60%
Nvidia's risk (cost of capital)	10.52% _> 8.49%	11.79% -> 8.50% (Higher riskfree rate + higher ERP)

With the changes made, and updating the financials to reflect an additional quarter of data,  you can see my Nvidia valuation in the picture below:

Nvidia valuation in January 2025 (Post DeepSeek)

There are two (unsurprising) results in this valuation. The value per share that I estimate for Nvidia dropped from $87 in September 2024 to $78 in January 2025, much of that change driven by the smaller AI chip market that comes out of the DeepSeek disruption (with the rest of the decline arising for higher riskfree rates and the equity risk premiums). The other is that the stock is overvalued, at its current price of $123 per share, even after the markdown this week. Since I found Nvidia overvalued in September 2024, when the big AI story was still in place, and Nvidia was trading at $109, $14 lower than todays price, estimating a lower value and comparing to a higher price makes it even more over valued..

    More generally, the value effect of the DeepSeek disruption will be disparate, more negative for some companies in the AI space than others, and perhaps even positive for a few and I have attempted to capture those effects in the picture below, comparing DeepSeek to a bomb, and looking at the damage zones from the blast:

In my view, the damage, in the near and long term, from DeepSeek will be to the businesses that have been the lead players in building the AI architecture. In addition to Nvidia (and its AI chip business), this includes the energy and gas businesses that have benefited from the tens of billions spent on building AI data centers. It is not that they will currently contracts, but that it is likely that you will see a slowing down of commitments to spend money on AI, as companies examine whether they need them. More companies are therefore likely to follow Apple's path of cautious entry than Meta and Microsoft's headfirst dive into the AI businesses. As for the businesses that are aiming for the AI products and services market, the effect will depend upon how much these products and services need data and computing power. If the proposed AI products and services are low-grade, i.e., they are more rule-based and mechanical and less dependent on incorporating intuition and human behavior, the effect of DeepSeek will be significant, with lower costs to entry and a commoditized marketplace, with lower margins and intense competition, If on the other hand, the AI products and services are high grade, i.e,, trying to imitate human decision making in the face of uncertainty, the effects of the DeepSeek entry are likely to be minimal and perhaps even non-existent. Thus, I would expect a business that is working on an AI product for financial accounting to find its business landscape changed more than Palantir, working on complex AI products for the defense department or commercial businesses. There is a grouping of companies, primarily big tech firms with large platforms, like Meta and Microsoft, where there may be buyer’s remorse about money already spent on AI (buying Nvidia chips and building data centers) but the DeepSea disruption may make it easier to develop low-cost, low-tech AI products and services that they can offer their platform users (either for free or at low costs) to keep them in their ecosystems.

    When faced with a development that could change the way we live and work, it is natural, especially in the early phases, to give that development a catchy name, and use it as a rationale for investing large amounts (if you are a business) or pushing up what you would pay for the businesses in the space (if you are an investor). In my early piece on AI, I talked about four developments in my lifetime that I would classify as revolutionary – personal computers in the 1980s, the internet in the 1990s, the smartphone in the first decade of the twenty first century and social media in the last decade, and how each of these started as catchall buzzwords, before investors and businesses learned to discriminate. Cisco, AOL and Amazon were all born in the internet era, but they had very different business models, and as the internet matured, faced very different end games. I hope that the DeepSeek entry into the AI narrative, and its disparate effects on different businesses in this space, will lead us to be more focused in our AI conversations. Thus, rather than describe a company as an AI company or describe the AI market as “huge”, we should be more explicit about what part of the AI business a company fits into (architecture, software, data or products/services) and apply the same degree of discrimination when talking about AI markets. If you also buy into my reasoning, you may want to follow up by asking whether the AI offering is more likely to fall into the premium or commoditized grouping.

The Bottom Line

    My early entry into Nvidia and my holdings of many of the other Mag Seven stocks have allowed me to ride the AI boom, I have remained a skeptic about the product and service side of AI, for much of the last two years. I can attribute that wariness partly to my age, since I cannot think of a single AI offering that has been made to me in the last two years that I would pay a significant additional amount for. I see AI icons on almost everything that I use, from Zoom to Microsoft Word/Powerpoint/Excel to Apple mail. I must admit that they do neat things, including reword emails to not only clean up for mistakes but change the tone, but I can live without those neat add-ons. Since I work in valuation and corporate finance, not a day goes by without someone contacting me about a new AI product or service in the space. Having tried a few out, my response to many of these products and services is that, at least for me, they don’t do enough for me to bother. In many ways, DeepSeek confirms a long-standing suspicion on my part that most AI products and services that we will see, as consumers and even as businesses, fall into the “that’s cute” or “how neat” category, rather than into the “that would change my life”, If that is the case, it has also struck me as overkill to expend tens of billions of dollars building data centers to develop these products, akin to using a sledgehammer to tap a nail into the wall. Every major innovation of the last few decades, has had its reality check, and has emerged the stronger for it, and this may the first of many such reality checks for AI.

    I know that much of what I have said here goes against the "happy talk" narrative about AI, emanating from tech titans and business visionaries. I know that Reid Hoffman and Sam Altman believe that AI will be world-changing, in a good way, relieving us of the pain of tasks that are boring and time consuming, and even replacing flawed "human" decisions with be more reasoned AI decisions. They are smart men, but I have two reasons for being cautions. The first is that I have had exposure to smart people in almost every walk of life - smart academics, smart bankers, smart software engineers, smart venture capitalists and yes, even smart regulators - but most of them have had blind spots, perhaps because they hang out with people who think like them. The second, and this perhaps follows from the first, is that I am old enough to have heard this evangelist pitch for a revolutionary change before. In the 1980s, I remember being told that personal computers would eliminate the drudgery of working through ledger sheets with calculators and pencils, but as young financial analysts will tell you today, it has just created a fresh and  perhaps even more soul-sucking drudgery, where monstrously large spreadsheets govern their workdays. In the 1990s, the advocates for the internet painted a picture of the world where access to online information would make us all more informed and wiser, but in hindsight, all it has done is weaken our reasoning muscles (by letting us look up answers online) and made us misinformed. In this century, social media too was born on the promise that it would keep us connected with friends, even if they were thousands of miles away, and happier, because of those connections, but as my good friend, Jonathan Haidt, and others have chronicled, it has left many in its orbit more isolated and less happy than before. 

YouTube Video

Nvidia Valuations

1. Nvidia valuation in September 2024 (Pre DeepSeek)
2. Nvidia valuation in January 2025 (Post DeepSeek)