Monday 21 December 2015

Can Prospect Theory Explain High Start-up Valuations?

Human beings are not particularly good at thinking about probabilities. The last several decades of research in psychology and behavioral economics have unearthed an array of cognitive biases in how we reason about uncertain events. For example, we are prone to misinterpret the results of diagnostic tests, by failing to account for the base rate of a disease in the population. This has enormous implications in the medical field, and may be leading us to over-diagnose and over-prescribe treatments for a variety of illnesses.

One of the foundational theories in behavioral economics is prospect theory, formulated by Daniel Kahneman and Amos Tversky. This theory is most well known for the observation that humans interpret losses as more consequential than gains of the same magnitude. Thus, simply changing the framing of a decision from a ‘gain frame’ to a ‘loss frame’ can make people much more averse to taking risks. In this blog post I want to focus on another prong of prospect theory: the over-weighting of rare events.

Prospect theory suggests—and experimental evidence supports—the idea that people ‘filter’ probabilities: we act as though very low probability events are more likely than they really are (and as though high probability events are less likely than they really are). This helps explain why people fret so much over low probability dangers such as shark attacks and ignore more mundane risks such as accidental falls. It also helps explain why people gamble money on lotteries even when the odds of winning are very slim, and the expected return is negative.

What has this got to do with valuing a start-up? The conventional way to value a company is to make a forecast of its future cash flows, then discount these back to find the ‘net present value’ of its future income. Alternatively, as a heuristic we can apply a multiplier to its earnings based on accepted valuations of other companies. Neither of these works for an emerging venture with a novel business proposition (i.e. your typical Silicon Valley start-up). The future prospects of such a company are shrouded in uncertainty.

Instead of trying to establish the likely path of a given venture’s future cash flows, investors—usually venture capitalists—take a portfolio approach. They pick companies they think will have a chance at becoming massively successful, but realize that many will fail to do so. Each investment is a bit like a lottery ticket. In the classical VC portfolio model, roughly one investment in ten would need to exit at a blockbuster valuation for the overall fund to make a decent return on investment.

In the present wave of technology venture activity, three key things are being done differently to the past. First, the definition of a ‘massively successful exit’ has inflated: ventures now aspire to be ‘unicorns’ with a billion-dollar valuation. Second, investors are spreading their money out, investing in a larger number of ventures. This is most visibly true in accelerator programs, which provide large numbers of nascent ventures with seed funding and mentoring in return for a small equity stake: they explicitly rely on a scattershot approach. Instead of a VC picking ten investments and hoping for two or three large exits, the accelerator approach is to invest in a hundred startup teams and hope for one unicorn. Third, more ventures are staying private for longer, rather than go public through an IPO. As described in this FT article, this allows them to effectively manage their headline valuation figure by giving new investors guaranteed financial returns (risking, in the process, the equity of preceding investors). This prevents negative opinions of the venture’s prospects from being incorporated in its valuation.

And so we have a perfect storm in which valuations are based on someone’s estimate that a given venture will become a unicorn, and—according to prospect theory—they are biased to overestimate how likely this is. For every thousand startups, maybe one of them will be hugely successful, but all of them might be valued as if they have a one-in-a-hundred chance of this success. This is a problem. More fundamentally, we are dealing with such small probabilities that we can easily get them very wrong.

Earlier in the year I considered a few possible mechanisms by which a hypothetical technology bubble might burst. Here, I’ve described one psychological factor that might be behind high startup valuations in the first place. It’s also worth noting that prospect theory can explain rapid changes in investor sentiment. If prices start falling—for example if a bubble shows signs of bursting—investors can switch from a gain mindset to a loss mindset, and immediately become much more risk averse. I hope this doesn’t happen, because the present wave of entrepreneurial activity is generating a lot of innovation. But a wise investor or entrepreneur should be aware that the tide might turn in the near future, and plan accordingly—or risk getting swept away when it does.

Saturday 31 October 2015

Let Them Eat (Micro)Chips: The Second Machine Age and the Spectre of Technological Unemployment

We are in the midst of the greatest economic upheaval since the industrial revolution. This is the premise of The Second Machine Age by Erik Brynjolfsson and Andrew McAfee, a book discussing the economic implications of present day technological trends. It is an excellent piece, which touches on several topics I have previously explored in this blog, from the trends towards scalability and the consequent ‘winner takes all’ market dynamics, to the deep challenges the information age poses to the measurement of economic growth.

The book has a compelling overarching theme: technology is driving two forces, one positive
for society and one negative. On the one hand technological change is generating an enormous bounty of economic growth. On the other hand, it is also driving increasing spread between rich and poor, and these economic faultlines could undermine the basic fabric of society.

Behind both bounty and spread is the rise of machine intelligence. Machines can take on ever more tasks, even ones that a decade ago we thought would be impossible to automate. The poster child for this is driverless cars. Technology experts used to think driving is so complex that humans would always have an advantage over computers, but the exponential progress of technology has rendered this prediction wrong. Google has been testing driverless cars for several years and Tesla’s Autopilot mode has already made automated vehicles a commercial reality.

One of the discussions in the book I find most compelling is on the subject of technological unemployment. At least since the days of the Luddites, the spectre of machines taking our jobs has worried generations of workers and commanded much attention in social and political science. The prevailing wisdom in the contemporary economics establishment is that technological unemployment is, indeed, a phantom, one we need not worry too much about. The argument goes as follows: while technological change may tear down old industries, it opens up new possibilities, and through the process of entrepreneurial action old ‘factors of production’ can be redeployed to productive uses. People whose skills become obsolete can learn new skills, they just need to be flexible about the type of work they are willing to do.

Brynjolfsson and McAfee make a compelling case that technological unemployment is a legitimate concern. They point to three main reasons:

1.) Rates of change
The argument against technological unemployment rests on the idea that people can adapt and find employment in growth industries. This reasoning holds as long as the rate of adaptation is faster than the rate of technological change. Historically this has been the case: despite the gales of creative destruction blowing strongly, society has adapted.* However, Brynjolfsson and McAfee point out that just because a trend held for 200 years doesn’t mean it holds forever. The rate of technological change has been increasing; can we expect that individuals and the institutions of society will adapt at an ever increasing rate as well?

2.) Elasticity of consumption
Another part of the argument against technological unemployment is that gains in productivity lead to lower prices, which in turn stimulate a higher volume of consumption. This assumption – that in aggregate the long-run “elasticity of demand” is approximately one – would provide an adjustment mechanism if technology continues to raise productivity. The authors point out that if this assumption is wrong, then economic growth would eventually come grinding to a halt.

A corollary of the elasticity of consumption argument, not addressed by the authors, is that it relies on prices going through periods of deflation. Deflation occurred in 1930s US and 1990s Japan, and might have occurred in the 2008-2012 Great Recession if it were not for the unconventional monetary policy of central banks around the world. Avoiding deflation has been celebrated for averting a potentially disastrous depression. But if prices are never allowed to fall, we lose one of the economic mechanisms for adjusting to technological change. I don’t have a clear cut answer to this dilemma, but I would like to see a few more economists discussing this issue.

3.) Floor on wages
This argument is presented with a thought experiment: 
“Imagine that tomorrow a company introduced androids that could do absolutely everything a human worker could do, including building more androids. There’s an endless supply of these robots, and they’re extremely cheap to buy and virtually free to run over time.” 
In this hypothetical end-game scenario, the equilibrium wage for human labor falls to zero. Managed well, we’d be in a Utopia, managed poorly – dystopia.

This is no mere parlor game. The authors also point out that in a digital economy, in which the output of superstars can be freely reproduced, the equilibrium wage for non-superstars is already zero, or close to it. These non-superstars look for work in other sectors, pushing wages down there. Eventually, “If neither the worker nor any entrepreneur can think of a profitable task that requires the worker’s skills and capabilities, then that worker will go unemployed indefinitely.”

Something implicit and horrifying in this last mechanism is that the free market’s solution to an oversupply of workers is starvation. When demand for a typical material good shrinks, its price falls. The supply side of the market adjusts by producing less of it. But when demand for labor shrinks, its price (i.e. the wage rate) may fall, but this doesn’t translate to a lower supply of labor (i.e. a lower population), except through violent means.

The book concludes that governments need to take action in order to make the most of technology’s bounty while minimizing the spread, or at least mitigating some of its worst consequences. Amongst other things, the authors advise greater investment in education (“higher teacher salaries and more accountability”), more support for entrepreneurship, more investment in science, and more progressive taxation, especially raising taxes on those with superstar levels of income.

The authors join a growing minority of commentators who view a Basic Income – a guaranteed minimum income paid by the government – as the best solution to rising inequality.** In the long run this might be the only feasible way to organize an economy which only requires a small, skilled minority to generate most of its economic output.

Technological change bears the potential to benefit everyone, but it also has the potential impoverish the majority while enriching a few. This is not some science fiction future – this is starting to happen today. The warning bells are sounding and now some wise leadership will be needed to steer the ship through the coming storm.


* Though not without spells of pain and suffering along the way.
** I find it quite noteworthy that this possibility is now part of mainstream discourse, and I found it even more surprising to learn that it almost became a reality under the Nixon administration

Wednesday 4 February 2015

How Might the Tech Bubble Burst?

Tech company valuations are through the roof right now, and many people have been questioning whether we are presently living through a tech bubble. In this blog post I set to one side the thorny question of whether we are in a bubble or not. Instead, I go through a thought experiment: I assume that we are in bubble, and play out a few scenarios for ways in which it might burst. Here are my top three:

Scenario 1: A Rising Star Falls
Present tech valuations are only warranted if you believe in the fundamental quality of their management teams. A high level of future growth is already priced into current valuations. This will only be attainable if they can expand their horizons, for example by expanding internationally. Such expansion puts a lot of pressure on organizational infrastructure. It is hard for a Silicon Valley-based headquarters to ensure that every local subsidiary maintains the quality standards they aspire to globally. Businesses who scale up slowly often have problems; those doing it at an accelerated pace (such as Uber) are even more likely to trip up. If too many local scandals mount, the managerial quality of the whole enterprise will get called into question, as will its valuation. The bold, fresh-faced management team will suddenly look hapless and inexperienced, flailing in the midst of a crisis, or riven by internal politics.

We only need look at Enron to see the risks that free-wheeling growth can expose an organization to. Now I am categorically not saying that every tech start-up is an Enron.  waiting to happen. What I am saying is that it could take just one high-tech corporate implosion to cast doubt on all the others. And once investors start to doubt the fundamental managerial quality of these tech ventures the game is up for the whole pack. Let me be totally clear: this line of argument is about perceptions. In the context of venture capital investments ‘risk’ is highly subjective, in other words it is a matter of perceptions. One dramatic fallen star could change the perceptions of investors about the risk of all the other stars, leading to the tech bubble deflating.

Scenario 2: The Advertising Pyramid Collapses
Many tech ventures rely on advertising as their main (or sole) source of revenue. Advertising is the bedrock of the tech sector, worth an estimated $43bn in 2013. It has allowed the industry to evolve in such a way that consumers expect services to be free. Take it away and those apps and websites suddenly don’t look like such appealing investments anymore.

Where do the pressure points lie when it comes to advertising? I see two potential sources of strain. First is the question of the proportion of advertising accounted for by tech companies and websites themselves. Apps tend to display adverts for other apps; websites display adverts for other websites. Webmasters buy ads to drive eyeballs to their site, where they hope people will click on ads. When this kind of behavior occurs, the online sector is feeding on itself – it is autophagous. Here we are in classic bubble or pyramid territory. The pyramid is sustained as long as it draws more people in to play the game, but once it is revealed to be hollow, it vanishes at once. I don’t have data on how much advertising is of this nature – so I can’t truly judge the extent to which this is a problem. However, just from my personal experience of browsing the web it appears that a lot of advertising is of this autophagous nature.

The second pressure point is the difficulty of measuring the return on investment (ROI) of online advertising. Web-based advertising platforms throw off a lot of data. The funnel from views-to-clicks-to-purchases can be tracked, so in principle a marketing manager can attribute a given online sale to a given online advert. However, things are far from simple. A customer who clicked through a pay-per-click advert may have still made exactly the same purchase even if the advert hadn’t been there. And a lot of online advertising is bought primarily to promote offline sales (think of, e.g., car adverts). However the effects on offline sales are much harder to track. Interestingly, while online advertising opens up the possibility of using randomized experiments to measure the effect of adverts, research so far has found that the effect size is so small it is hard to reliably measure from a statistical standpoint.

The upshot of this: the online advertising industry, while huge, is not yet in a long-term, stable equilibrium, and it’s not clear whether the stable market size will be larger or smaller than the market that exists now.

Scenario 3: Silicon Valley Disrupts Itself
To disrupt an industry is the bold aim of many of Silicon Valley’s start ups. It typically entails finding a way to deliver the same service the industry presently delivers but at a fraction of the cost or at a step-change improvement in quality or convenience.* It is often said that to disrupt you need to offer something 10x better than what presently exists, in order to overcome people’s inertia and lock-in with present systems.

The industries targeted for disruption are typically of the staid, old-school variety, perhaps dominated by some entrenched rentiers – think of Uber disrupting the taxi / car industry, or TransferWise disrupting the foreign remittance industry. The narrative of disruption underlies the massive valuations these companies receive. To take Uber, for example, in debates about its valuation have moved from comparisons with the entire global market for taxi services to discussion about how it might act as a subtitute for car ownership.

But there's no particular reason only old industries are vulnerable to disruption. It is perfectly conceivable that one of the current tech giants itself gets disrupted. People appear pretty locked-in to social media platforms such as Facebook, but if a company were to come along with an offer 10x better than one of these, it is easy to see customers switching. And, in fact, many already did: in the last few years plenty of people switched from Facebook to Twitter or Instagram as their primary social media feed, and in future something even better than either of these may come along. Well aware of this fact, Facebook paid a billion dollars for Instagram and $22bn for WhatsApp primarily because of the threat that these businesses posed to its dominance.

I can’t predict what a social media platform 10x better than Facebook would look like, but then neither could I have predicted Facebook’s creation before it existed. And, as with Scenario 1, it only takes one giant to fall asunder for many of the others to lose their appeal to investors.

So, there we have it, three scenarios for how the tech bubble might burst. Which do you find the most plausible? What other scenarios sounds realistic? Answers in the comments below! 


*Note: this is a fairly colloquial definition of disruption. It is a somewhat warped version of its original academic usage by Clay Christensen, which referred to creation of new performance dimensions for emerging market sub-segments that eventually become large markets in their own right. Here, the colloquial meaning is the one I intend.

Friday 16 January 2015

Notes from the cutting room floor: "Google has mastered technology, but they need to better understand people"

I wrote the following paragraph on 3rd November, 2013, shortly after a big rise in Google's share price. I never got round to expanding it into a full post. This week the Google Glass developer program was put on hiatus so it seemed like an apt point to dig it up. 

Google has made the news this week for its share price reaching record highs. I’ve been observing the company keenly for many years and greatly admire the technical prowess and creativity of its employees and the vision of its founders and leaders. In the past the company has pioneered a vast array of internet services and is one of the chief reasons why much of what is online is available for free (including this blog). Now it looks as though they plan to be a pioneer in the hardware world, and this is likely to be the greatest challenge they will ever face. There are numerous possible miss-steps, and above all I want to highlight the risk that, while they have mastered technology, they lack an understanding of how it is socially adopted and accepted. This will leave some of their greatest innovations – Google Glass and driverless cars – down a rocky path, with a real chance of outright rejection by society.