Is the university system doomed?

David Perell’s essay “What the Hell is Going On” provides a thought provoking review of how technology and the information age continues to disrupt the most time honored information markets: advertising, education, and politics. Much of his discussion is inspired by the hypothesis of Martin Gurri that the information age has permanently diminished the authority of existing elites. Perell is willing to go out on a limb a bit in its quest to “solve for the equilibrium” (as Tyler Cowen would put it).

This includes the bold claim that the university system is doomed to be replaced by online courses and new types of signaling. According to Perell, the proliferation of low-cost and high-quality online education spells the end of universities, or at least the middling ones.

In his recent book The Case Against Education, Bryan Caplan lays out the opposite case, and it’s worth examining the differences. Caplan begins with the point that the correlation between college graduation and later income appear substantial, clocking in at 30% (correcting for ability bias) even for “impractical” majors like Philosophy and Art History, and much higher for hard sciences and engineering.

He notes that this link between education and income can be ascribed to a mix of three sources:

  1. Social capital added. The “improvement” of the individual via training and coursework.
  2. Pre-existing skill. The tendency of the competent to enroll in college.
  3. Signaling. The tendency of employers to prefer candidates with better credentials.

Caplan argues that signaling accounts for ~80% of the premium for college graduates. He provides research to back that number up, but he also asks a question that’s hard to ignore: “Would you rather have 4 years of world-class education at Princeton and no degree, or a Princeton degree without the coursework?” I for one am fairly convinced by the argument that the latter would make you better off financially than the former, even if you believe that the college experience itself is a priceless spiritual journey.

Bryan goes on to argue that the signal provided by a degree is valuable to employers, because it connotes a package of three traits: intelligence, conscientiousness, and conformity. A technology company hiring a philosophy major as a project manager isn’t counting on them to produce well-argued treatises on Locke, or make use of any information about philosophy whatsoever. Rather, the company is counting on them to be able to reason well about novel situations (intelligence), ensure that their work is produced on time and to a high standard (conscientiousness), and that they will “get with the program” and act in accordance with company values and incentives (conformity).

If college degrees are truly 80% signalling, then even the best online courses or MOOCs could not replace them. Under this hypothesis, universities do not owe their success in the modern world to their status as gatekeepers of information. Rather, owe it to their reputation for educational programs whose completion signals intelligence, conscientiousness, and conformity. As Bryan points out in later chapters, the conformity signal in particular provides a catch-22 to would-be disruptors: Doing anything other than going to a four year college would be non-conformist.

So what’s the prognosis right now? There are predictions circulating that the demand for 4-year colleges will diminish starting around 2026, but these have to do with a drop in the number of high school graduates rather than a reduction in the proportion of graduates attending. Perell claims that mid-sized liberal arts schools are already beginning to decline, but it’s not clear that the institution more broadly is suffering.

I don’t doubt that there is opportunity in the education space: college provides little in the way of useful job skills, and as a signaling device, it is both costly and inequitable. Delivering content and university-like interactions should be feasible in an online format. But delivering on the strong signal currently sent by attending a selective four-year institution, and especially signaling conformity, is an uphill battle for any disruptor. We’ve already seen some experiments that have produced quality content but failed to make a dent against 4-year degrees (e.g., Khan academy, edX) and a generation of boot-camps that have been hugely useful to a niche set of students, but may not be able to scale to the level of universities (e.g., Lambda School, App Academy). There’s still an opening for new entrants, and for an ensemble of these disruptors to coalesce into a better route to success for new high school graduates. But the university system is a hardy weed with deep pockets and a centuries-long track record. It won’t be uprooted overnight.

Correction 2019-02-10: I originally had listed Philosophy’s premium as ~20%, in fact Caplan lists it as closer to 30% after correcting for ability bias. I also added the graph of adjusted premia.

Bibliography

Caplan, Bryan (2018). The Case Against Education: Why the Education System Is a Waste of Time and Money. [amazon]

Fox, Justin (2019). “The Coming College Enrollment Bust.” Bloomberg Opinion. [link]

Gurri, Martin (2018). The Revolt of The Public and the Crisis of Authority in the New Millennium. [amazon]

Perell, David. “What the hell is going on?” [link]

Perell, David & Forte, Tiago. “Tiago Forte: The Future of Education.” [link]

David Perell on the Paradox of Abundance

David Perell recently published a long read article on the new age of democratized digital media we are entering. Although he gives a history of the last century of media, his ultimate message centers around how we should be consuming information in the digital era. The most important concept he shares is the Paradox of Abundance. Briefly, we have access to more information than ever before. Because there is so much information out there, information purveyors are in a desperate zero-sum competition for eyeballs. They respond by creating urgency and emotional hooks around articles and information that aren’t actually very important. This confronts consumers with a tsunami of clickbait and emotional appeals, which either sends us running from false crisis to false crisis, or makes us check out from media altogether. This means that people who can focus on truly useful and enlightening content are richer than ever before, while the median user is worse off, tugged in all directions by emotionally captivating but ultimately non-enriching content.

This feels most real to me whenever I log onto Twitter. The transcendent moments on Twitter are truly transcendent. Rarely do I spend 20 minutes on Twitter without encountering some new perspective or piece of information that I hadn’t considered before. But the tsunami can be overpowering: I often spend far longer than I had planned following trails of seemingly alarming scandals or political trends. Twitter’s algorithm accelerates this, bubbling the most alarming, shocking, or polarizing content to the top of the feed. My Feedly experience could not be more different: Since I mostly follow only academic blogs or writers I deeply respect, my feed is a bit more boring but also has a much higher signal-to-noise ratio. It’s also less addictive.

Perell draws an analogy to eating right that feels particularly apt. Ultimately this may be the sort of thing our culture needs to adapt to. Just as abundance in food resulted in the evolution of cultural norms around eating right, we may need to develop cultural norms around “informing” right. My feedly is more like a quinoa bowl, whereas a twitter binge can leave me feeling like I just ate a bagful of tootsie rolls. He even mentions writing as analogous to cooking at home, another regularizing function that can improve diet.

There is one question Perell leaves unanswered: What is it all about? Is there an individual duty around online engagement as a sort of civic service? Or would we all just be better off unplugging and reading books? Either way, it is clear that information abundance is here to stay, and passive consumption of whatever passes across our screens is no longer an acceptable strategy.

Is malinvestment in advertising killing D2C ecommerce?

Alex Taussig of Lightspeed writes that there is a typical playbook for vertical ecommerce brands: “(1) Launch a “hero” product that provides branded product innovation, (2) at high gross margins, (3) enabled by an owned & operated e-commerce experience, (4) with high marketing efficiency, usually driven by viral or word-of-mouth campaigns.” (Taussig 2019). The filings from Blue Apron and now CSPR Casper that item (4) on that list, marketing efficiency, can prove elusive as these brands attempt to scale to the level of the expectations they’ve set with investors.

When Blue Apron released their S-1, I was working on Stitch Fix’s own S-1, and so I immediately found their filing interesting. I looked into the numbers that I knew to be important for a subscription service: customer net present value and acquisition cost. From what I could glean, NPV on a customer basis actually looked fairly high; my guess was around $200 assuming a reasonable discount rate. But marketing costs were astounding: ~$144M in 2016, and with Q1 2017 marketing annualized to over $242M. We can’t say for sure how many new customers they paid for with that. If we use the difference in number of customers at the beginning of the period and the end (this is conservative due to the effect of churn), the implied blended CPAs are $320 in 2016 and $385 in Q1 2017. That already implies a money losing proposition, but it obscures how bad things really were. The marginal acquisition cost is typically much higher than the blended cost, because advertising does not scale up efficiently. It’s impossible to know exactly what the ratio is between blended and marginal. In my experience, 2x is a fairly conservative estimate. This means it is quite possible that at the margin, APRN was largely paying $700 or more to acquire a customer worth only around $200. When I saw that, I knew they were toast right away.

Now we have a new filing from Casper with a similar trend. Although Casper claims to establish deep customer relationships, they have a repeat rate of only ~14%, suggesting that we can safely treat marketing acquisitions as one-off. AOVs were $710 in the first nine months of 2019. Revenue plus discounts (which are included in AOV) were $392M, with COGS + refunds of ~$237M, for a profit margin of 39%, or $277 per order. Like APRN, this value per acquired customer is quite impressive, and may be slightly understated if customers are coming back unprompted to buy more big ticket items. Using the above numbers, we can assume 552k acquisitions on a sales cost of $114M, implying a blended CPA of $202. This may seem like a reasonable figure until, again, one considers that the marginal cost is likely to double the blended figure or more. In this case, again, this means that Casper may well be paying a $400 marginal CPA to acquire $277 of gross margin. This is not quite as dire as the Blue Apron situation, but far from comfortable. We can’t tell from this report how much value there is in those repeat orders, but they are absolutely relying on them to make their strategy work.

Direct to consumer brands succeed by providing good margin on product via vertical integration and efficiencies in customer acquisition via strong branding. Often marketing is easy in the early days, when word of mouth provides costless customer acquisition. But $APRN and $CSPR show that acquisition efficiency does not come for free for these brands, especially once they attempt to scale past the growth level supported by referrals. Once organic and word of mouth peter out, the temptation is strong to dump money into paid acquisition well past the point of efficiency. Casper has already had to reduce its offer price below its recent round; this week we’ll see the public market’s judgment.

Bibliography

Form S-1, Blue Apron, filing date 6/1/2017 [link].

Form S-1, Casper, filing date 1/10/2020 [link].

Taussig, Alex. “Drinking from the Firehose #154.” 1/13/2019.

How a dart-throwing monkey can beat the S&P 500.

Main points:

  • A monkey throwing darts can beat the S&P 500 on a risk adjusted basis.
  • This is because cap weighting has historically been a poor portfolio construction strategy.
  • Passive portfolios aren’t fragile. Random perturbations are generally benign as long as they tilt toward the so-called “smart beta” factors such as value and small cap.

I recently read a paper that changed the way I think about passive indexing.

Cap weighting (short for market capitalization weighting) has long been upheld as the gold standard of passive investing, from Bogle’s Common Sense on Mutual Funds to Warren Buffet’s famous bet that hedge funds could not beat the S&P 500. Market capitalization is the implied value of all of a given company’s shares if you added them up at the market price. Indexes like the S&P 500 weight the importance of stocks based on market capitalization. So at ~1.3T, Apple, would receive over twice as much weight as Facebook, whose capitalization is ~600M. In theory, if markets are efficient, a cap weighted portfolio is perfectly efficient because it reflects the allocation of market capital as a whole. If some other weighting of assets could produce higher returns, one would imagine that the market as a whole would shift into those assets, restoring equilibrium.

As it turns out, however, it is easy to outperform the index. It is so easy that a monkey can do it. Arnott et al (2014) investigate several alternative portfolio allocation strategies that have been proposed in the literature and appear to outperform cap weighting. Shockingly, they find that these strategies work just as well, or often even better, when they are inverted. This is strange enough that it’s worth spelling out clearly. In the investment literature, researchers commonly identify some set of characteristics that might indicate that a stock will produce outsized returns. It is common practice to test these hypotheses by looking at the simulated performance of portfolios constructed using those characteristics at different points in the past. For example, one might imagine that stocks with high 5 year average earnings would be strong performers. In fact, these stocks do have higher returns: Arnott et al. estimate 11.18%, as opposed to 9.66% for a U.S. cap weighted index.

The authors make an astonishing revelation by flipping the strategies on their head. For example, they invert 5 year average earnings strategy, simulating portfolios filled with the stocks that earned the least. Common sense implies that the inverse of a winning strategy should be a losing strategy. Instead, the inverse index actually performs better, with returns of 14.38% and an improved Sharpe ratio (the ratio of expected return to variance in return). The authors show similar results for about a dozen other strategies, using two different techniques for inversion. This is almost spooky: If a strategy is good, how can its opposite also be good?

Superficially, the answer is clear. What do the 5 year high earner portfolio and 5 year low earner portfolio have in common? They are both not cap weighted. Cap weighting is such a bad way of selecting stocks that it is beaten by almost any alternative strategy! To drive this point home, the authors simulate a dart-throwing monkey, picking portfolios of 30 stocks completely at random, and show that these portfolios also beat the cap-weighted index: The monkey portfolios produced returns of 1.6 percent in excess of the index, with a better Sharpe ratio (0.33 vs. 0.29). Looking more deeply at why this is the case, the authors show that most of the excess returns they observed across all these strategies are explainable by an enriched Fama-French model. This is a widely discussed model in the finance world that assumes that investors are compensated for certain types of exposure, which are known as factors. In this case, the portfolios were typically weighted toward value and small-cap stocks. These anomalies are often referred to as Smart Beta, because the Fama-French equation uses the symbol beta to denote the coefficients for the factors.

I’ve been concerned in the past that smart beta indexing is inimical to the philosophy of passive investing. Trying to outsmart the index exposes your portfolio to the precise stock-picking scheme used by a bespoke ETF. Because of this, I had always been suspicious of Betterment’s decision to tilt the portfolio toward value, and the recent underperformance of value seemed to bear out my bias. But if even a monkey can pick a good portfolio, it implies that the effects are quite robust to the precise method of stock picking so long as you’re not actively trying to weight against the smart beta factors.

The literature is not clear on exactly how or why smart beta outperforms the cap weighted index. This outperformance is enigmatic since it seems contrary to EMH. Some argue that they result from investor irrationality or the perverse incentives of fund managers (e.g., Baker et al. 2011), while Fama and French themselves seem to believe that this reflects efficient pricing of hidden risk factors (Fama & French 1993). There is good reason to doubt that the past performance of these factors might not be consistent in the long run, since the explanation is not clear. Nevertheless, this paper made me stop worrying and love the tilt.

Resources & Bibliography

Arnott et al (2014) “The Surprising Alpha from Malkiel’s Monkey and Upside-Down Strategies.” The Journal of Portfolio Management. https://thereformedbroker.com/wp-content/uploads/2014/11/jpm_summer2013_rallc.pdf

Arnott et al. (2016) “How Can ‘Smart Beta’ Go Horribly Wrong?” https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3040949

Baker, Bradley, and Wurgler (2011) “Benchmarks as Limits to Arbitrage: Understanding the Low-Volatility Anomaly” Financial Analysts Journal. https://doi.org/10.2469/faj.v67.n1.4

Bogle, John (1999). Common Sense on Mutual Funds.

Fama and French (1993), “Common risk factors in the returns on stocks and bonds.” Journal of Financial Economics.

Fama and Thaler discuss efficient markets, including the Fama-French 3-factor model. https://www.youtube.com/watch?v=bM9bYOBuKF4

What is a tech worker worth?

As a tech worker I’ve always been bewildered by the high compensation on offer for technical workers in the Bay Area. If you’re not in tech and you want to see the gory details, see this site, which is quite accurate as far as I can tell.

I’ve sometimes wondered if the whole thing is some sort of Ponzi scheme or bubble: Employers don’t want to feel like they’re getting “second best” labor, so the compensation rises past a level that might be reasonably justified on an ROI basis. The whole song and dance is funded by VC money and a few very lucky big ticket exits that keep the party going.

Academic economists would typically point to an efficient market perspective: Employers act in rational self interest, overpaying for labor would be a very expensive mistake, and so the prices probably simply reflect workers capturing a larger portion of their contribution. But could we really be worth so much?

It turns out we can actually answer this question. Dimmock et al (2019, pdf) were able to do a causal analysis using random assignment provided by the H-1B lottery (h/t Alex Tabarrok). When you hire a worker on an H-1B visa. There’s a lottery in Spring, and you have a 50/50 chance of winning. If you win, the visa takes effect in September. Otherwise you’re out of luck.

Maybe I shouldn’t have been surprised, but the results are extreme. Winning a single H-1B lottery raises the likelihood of a successful IPO dramatically. Each standard deviation increase in win rate (which was roughly equivalent to a single incremental win) increased the rate of successful IPOs within 5 years by 23% (rising from 6.6% to 8.1%). How much does that make each H-1B win worth? If the marginal benefit of a lottery win is approximately 1.5% of an IPO, and we assume an IPO within 5 years is worth $100M relative to alternative outcomes, then that worker is worth $1.5M. And what’s especially amazing is that this is simply value over replacement player: Typically, when a company loses the lottery, they re-open their hiring req and hire someone else. So this is actually the value of the immigrant worker over and above the domestic alternative.

I see two main takeaways here. One is, immigration is underrated. As Bryan Caplan has been shouting from the rooftops, immigrants don’t just “take our jobs,” they contribute in ways that are irreplaceable. Second, as heady as the market for tech workers may appear, it is not a bubble: At least in aggregate, tech workers are writing their own check.