You are here

Albert Wenger

Content Written by Author

Wednesday, October 18, 2017 - 11:30am

I ended the previous Uncertainty Wednesday post asking whether an expected value always exists. Going back to the definition, the expected value is the “probability weighted average of a random variable.” So let’s construct an example of a random variable which does not have an expected value. We will consider a probability distribution with infinitely many discrete outcomes, in which the first outcome has probability ½, the second ¼, the third 1/8 and so on. This is a valid probability distribution because all the probabilities sum up to 1:

½ + ¼ + 1/8 + 1/16 + …. = 1

Whether or not an expected value exists depends on what the numeric values of the outcomes are. Consider for instance the random variable where the first outcome is 1, the second is 2, the third is 3 and so on. For this random variable we have an expected value, because

EV = ½ * 1 + ¼ * 2 + 1/8 * 3 + 1/16 * 4 + …. = 2

Why is that so? Because even though our random variable includes ever larger outcomes, these very large outcomes occur with very small probability and so the probability weighted average is a convergent infinite sum.

But now consider what happens when the outcomes themselves grow exponentially. Let’s consider the case where the first outcome is 2, the second is 4, the third is 8 and so on. Now we have

EV = ½ * 2 + ¼ * 4 + 1/8 * 8 + 1/16 * 16 + …. 
EV = 1 + 1 + 1 + 1 …

Clearly the EV here is no longer a convergent sum but rather diverges towards infinity.

Now you might say, Albert that’s not an example of an expected value that doesn’t exist, the expected value is simply infinite. This might take us into a separate discussion of the meaning of infinity, which might be fun to have, including the more sophisticated objection to the example which would claim that all real processes have some finite upper bound.

For now though let’s focus on a different question: how does the sample mean behave for the random variable we just defined? This is a well defined question. A sample has, by definition, a finite number of observations (that’s what it means to be a sample). So each sample will have a mean. What is the implication of the expected value diverging for the behavior of the sample mean?

Monday, October 16, 2017 - 11:30am

I woke up this past Friday (the 13th) to a DM on Twitter saying that Continuations was down. I immediately tried to open the site on my phone and was greeted by an ominous:

At first I figured that maybe Tumblr was down. A quick check of other Tumblrs revealed that not to be the case. At this point a somewhat queasy feeling started to set in. After a quick shower I went to my laptop and tried to log into my account only to see a terrifying sight:

At this point I was in a full blown state of panic. I have been writing here for a long time and my last full backup of Continuations was several years old!

Thankfully I know several people connected to Tumblr and they kindly offered to help. Continuations was fully restored within a couple of hours, but those were some scary hours in which I kicked myself for not following the advice that I give every USV portfolio company, which is to make sure to backup of all their data.

I have since learned that Tumblr does an excellent job keeping data around making it easy for them to restore things after an accidental deletion (apparently some automated bot deletion system had malfunctioned and removed Continuations). Still, I am feeling much better now that I have a once again current backup of Continuations.

This experience has made me think about other cloud services that I use extensively such as Google and Dropbox. I am now wondering if I should back these up to each other for increased redundancy. I am curious to hear from anyone who does that as to why and how they have set it up.

Thursday, October 12, 2017 - 7:30am

Last Uncertainty Wednesday we dug deeper into understanding the distribution of sample means. I ended with asking why the chart for 100,000 samples of size 10 looked smoother than then one for samples of size 100 (just as a refresher, these are all rolls of a fair die). Well, for a sample of size 10, there are 51 possible values of the mean: 1.0, 1.1, 1.2, 1.3 … 5.8, 5.9, 6.0. But with a sample size of 100 there are 501 possible values for the mean. So with the same number of samples (100,000) the distribution will not be approximated as closely. We can fix this by upping the number of samples to say 1 million. Thanks to the amazing speed of a modern laptop even 100 million rolls of a die just take a couple of minutes (this still blows my mind). Here is the resulting chart:

Much smoother than before! We could make that even smoother by taking up the number of runs even further.  

OK. So what would happen if we went to sample size 1,000? Well, by now this should be easy to predict. The distribution of sample means will be even tighter around 3.5 (the expected value of the distribution) and in order to get a smooth chart we have to further up the number of runs.

So what is the limit here? Well, this get us to the law of large numbers, which essentially states that the sample average will converge to the expected value as the sample grows larger. There is a strong version and a weak version of the law, a distinction which we may get to later (plus some more versions of the law).

For now though the important thing to keep in mind is that when we have small sample sizes, the sample mean may be far away from the expected value. And as we see above that even for a super simple probability distribution with 6 equally likely outcomes there is considerable variation in the sample mean even for samples of size 100! So it is very easy to make mistakes from jumping to conclusions on small samples.

Next Wednesday we will see that the situation is in fact much worse than that. Here is a hint: every sample has a mean (why?) but does every probability distribution have an expected value

Tuesday, October 10, 2017 - 7:35am

I am excited about a new open source project called idyll. Here is how Matthew Conlen, the lead author, describes idyll

Idyll is a tool that makes it easier to author interactive narratives for the web. The goal of the project is to provide a friendly markup language — and an associated toolchain — that can be used to create dynamic, text-driven web pages.

Idyll helps you create documents that use common narrative techniques such as embedding interactive charts and graphs, responding to scroll events, and explorable explanations. Additionally, its readable syntax facilitates collaboration between writers, editors, designers, and programmers on complex projects.

The project seems like an important step in the direction of an interactive learning environment that seamlessly combines text, mathematical formulas, code, graphics. Creating such an environment and then using it to share knowledge about the consilience of math, physics, computation and more is one of my three passion projects.

An example of an idyll document explains the etymology of the trigonometric functions. In a future version of idyll it will be easy to show and even edit the code behind the unit circle graph on the right.

If you are as excited about idyll as I am, please help me support the project via the idyll Open Collective page.

Wednesday, October 4, 2017 - 11:30am

Last Uncertainty Wednesday, we started to look at the behavior of the mean of a sample by repeatedly drawing samples. We used a sample of 10 rolls of a fair die. We know that the expected value of the probability distribution is 3.5 but we saw that the sample mean can deviate substantially from that on a small sample. In particular, with 10 rolls we got sample means both close to 1 (almost every roll is a 1) and close to 6 (almost every roll is a 6).

The fact that the sample mean itself is random and has a distribution shouldn’t be surprising and yet it is the source of a great deal of confusion. Let me show that in the case of discussion of weather and climate. I had defined climate as the probability distribution of possible weather events. The realized weather then is a sample. So we should not at all be surprised to see variability in the weather relative to past averages. And yet we use terms such as “unseasonably” cold or “unseasonably” hot all the time, which imply that there is something out of whack with what was observed. The challenge then in analyzing climate change based on data is to separate variability within the existing distribution (climate) from changes in the distribution (climate). We will get back to that in future posts, but first we have more on sample means.

What happens is we make our sample larger? Instead of a sample size of 10 rolls, let’s consider a sample size of 100 rolls. Below are graphs contrasting the results of 100,000 runs for sample size 10 and sample size 100:


We again see a distribution in the sample mean but it is much tighter around the expectation of 3.5, with almost all observed sample means for size 100 falling between 3 and 4, as compared to 2 and 5 for sample size 10.

It is really important to let this all sink in deeply. Even at a sample size of 100 rolls, there is significant variation in the sample mean. The good news is that the distribution of the sample mean is centered on the expected value. This is often referred to as the sample mean being an unbiased estimator of the expected value. We will dig into when and why that’s the case as it is not true for all underlying probability distributions (almost certainly *not* true for weather). 

The bad news though is that even when the sample mean is an unbiased estimator of the expected value, on any one sample that you draw if it is the *only* sample, you have no idea whether you are above or below the expected value. Keep in mind that all this analysis we are currently conducting is based on a known distribution. That is hardly ever the problem we actually confront. Instead, we have explanations which lead us to prior beliefs about distributions and we need to use the observations to update those beliefs

More to come on sample means and what we can learn from them next Wednesday. Until then, here is a question to ponder: why did the graph for sample size 10 comes out smoother than the one for sample size 100?  

Tuesday, October 3, 2017 - 7:30am

I was going to write a post today about how liquidity in financial markets goes down as concentration of holdings goes up but then I woke up to the news of the mass shooting in Las Vegas last night. I have written extensively before about the need for better gun control, so won’t rehash that today other than to note that the statistics for 2017 show already over 11,000 deaths year-to-date.

What yesterday’s mass shooting does make eminently clear yet again though is just how much damage and trauma a single person can inflict using modern technology. It is so much easier to destroy a life than to build one. Split seconds of pulling a trigger, versus decades of nurturing and growth. This fundamental asymmetry is one that we as humanity need to pay more attention to as we make more technological progress. 

The asymmetry between destruction and creation will never go away. It is baked deeply into the fabric of reality. There are myriads of arrangements of the molecules found in a human body and only a tiny fraction of those arrangements amount to a person who is alive and well. So as we have more and more power at our disposal we need to think carefully about how to prevent ever more destruction brought about by individuals (and small groups). Unfortunately, there are no simple answers here and we will be forced to look at uncomfortable trade-offs.

Saturday, September 30, 2017 - 7:30am

I just spent a couple of days in Vienna, Austria at a conference. The upcoming Austrian elections were the talk of the town. One of the major parties is campaigning in part on the promise of introducing an estate tax for all estates about 1 million Euros. Austria has a wealth distribution that is also highly concentrated with the top 1% of the population controlling 40% of total wealth.

Also during the last couple of days the new tax proposal by the Trump administration includes an elimination of the estate tax, despite the current exemption being already more than $5 million. Put differently, only estates above $5 million pay estate tax. The argument that this forces millions of farms and small businesses to be sold has been thoroughly debunked as most of them fall below this threshold.

To me the most fascinating aspect of the estate tax debate is how it relates to the justification for a universal basic income (UBI). One of the objection to a UBI is: what have people done to deserve it? The answer of course is that people don’t need to have done anything to deserve it. We can afford UBI because we collectively have inherited the technological, social and economic progress made by the generations that came before us.

Often the very people who want to eliminate the estate tax are the ones bringing up the “but they don’t deserve it” objection to UBI, seemingly oblivious to the complete logical contradiction this entails. So: if you believe that anyone deserves to inherit significant money from their parents, then you have no standing question whether people deserve a UBI.    

Thursday, September 28, 2017 - 5:10pm

Last Uncertainty Wednesday I introduced a definition of climate as the probability distribution over weather states. That post ended with a question about how historically observed statistics relate to this distribution. I am in Vienna Austria for a conference and the temperature yesterday got up to 70 degrees (Fahrenheit). Intellicast shows historic average high temperature for September 27 to be 65 degrees (unfortunately not saying how many years of data are averaged which is something I will try to track down). So clearly yesterday’s high is above the average high observed for this day in the past. But what should we conclude from that?

To get there we will first learn a bit more about the relationship between the expected value of a distribution and the observed sample means. As I had pointed out in my original post on expected value, this is the source of a great deal of confusion.

Rather than the complexity of weather, let’s take a really simple probability distribution: rolling a fair six-sided die. The probability distribution is really simple. Each of the values 1, 2, 3, 4, 5 and 6 has equal probability of 1/6. Hence

EV = 1/6 * (1 + 2 + 3 + 4 + 5 + 6) = 1/6 * 21 = 3.5

Now let’s look at a variety of samples from this distribution and the observed sample mean. To look at this I wrote some hacky Python code which you can see here (which I stored in a file called

from __future__ import division
from random import randint
import sys

runs = int(sys.argv[1])
dist = {}

size = int(sys.argv[2])
total = 0

for run in range(0, runs):

   total = 0
   for i in range (0, size):
       r = randint(1, 6)
       # print i+1, “: ”, r
       total += r
       # print “Sample mean: ”, total / size

   mean = total / size
   if mean in dist:
       dist[mean] += 1
       dist[mean] = 1

for mean in sorted(dist):
   print “%s: %s” % (mean, dist[mean]);

The inner loop creates a sample of size provided in the second command line argument and computes the mean. The outer loop runs that as many times as provided in the firs command line argument. So python 1000 10 for instance takes 1,000 samples of size 10 each and computes their respective means and counts how many time each mean occurs.

Here is the distribution of sample means for sample size 10 that results from that for different numbers of runs (I took the output from the above program, pasted it into Google Sheets to graph it):


What is going on here? What we are seeing is that the sample mean for a small sample of size 10 can vary a lot. In fact in 100,000 runs we see sample means very close to 1 (meaning most rolls came up as a 1) and some very close to 6 (meaning most rolls came up as a 6). As we graph the sample means for few runs they come out with a lot of their own variance. But as we draw a lot of them they seem to have the shape of a normal distribution.

Next Wednesday we will dig deeper into this phenomenon. And we will also look at what happens when we increase the size of our samples.

Tuesday, September 26, 2017 - 5:10pm

Having been outspoken against Trump since before the election, I am encouraged to see that we may have found a particularly effective form of protest in #TakeAKnee (or #TakeTheKnee). We have the initiative of Colin Kaepernick and those who followed his lead to thank for that. I realize that this started out as a protest against police violence as part of the Black Lives Matter movement during the Obama administration. I believe it has the potential though to morph into a broad and effective protest against Trump without losing its important origin.

One of the particularly frustrating aspects of watching the ascendancy of Trump was how impervious he was to any rational critique. In fact, his campaign was designed to obliterate such an approach by embracing outright lying and outrageous statements from the get go. Trump was extremely effective at labeling individual opponents in ways that made them seem weak. He was aided in that online by a large meme creation army.

But now Trump has found an adversary that is of a different nature: a symbol that can spread and be adopted by anyone. TakeAKnee has the potential to be the first truly successful Trump protest meme. And like other memes it is antifragile. The more Trump and his supporters attack it, the more media time it receives and the more it spreads.

What makes it so powerful is that it appears Trump is taking TakeAKnee as disrespectful of him personally (his rhetoric about flag or anthem notwithstanding, everything with Trump is persona). So when he sees it he can’t help himself but react and tweet, which then gives it coverage, gets more people to participate and sets off another round. Trump is easy to provoke. This is one of the attributes that makes Trump so inappropriate as a leader and is playing itself out with potentially terrible consequences vis a vis North Korea. TakeAKnee aims straight at that weakness.

So with this in mind, here is a cartoon that captures the potential. Exercise your right to free speech, help protest violence against black lives, and give Trump more opportunity to show his true nature by sharing this meme and if you have the opportunity participating.

Monday, September 25, 2017 - 11:30am

By now most CEOs of startups understand that they urgently need to figure out what role machine learning will play in their business. Large established companies too are actively engaged in this process. This is not an easy task in and of itself, as machine learning isn’t a panacea for everything wrong with your business and you cannot just sprinkle it on top of your existing business process and strategy. Instead, you likely have to jettison many assumptions about “how things are done” in your industry.

Suppose you have identified a genuine opportunity to apply machine learning, the next obvious challenge becomes a question of how to pursue it. Should you build something on your own or should you buy from a vendor?

Here I believe leaders on the product and business side are not always getting great advice from their engineering departments. Why? Because building machine learning systems from the ground up in-house is what every engineer wants to do. What could be more exciting for an engineer today than getting to build a machine learning system using TensorFlow? And yet in many instances that will be the wrong thing to do compared to using a specialized service provider.

Why? Because while it is easy to get going and achieve moderate accuracy, it is quite difficult to build something that improves in accuracy over time and delivers high availability and low latency at scale. So you should approach the decision of what to build and what to buy with the same clarity as you would in other areas. For instance, does it make sense for you to run your own database instances and patch them, upgrade them, scale them or should you use a managed database service? Not so long ago everyone ran their own databases but today even large enterprises are shifting to managed services.

The same is true for machine learning. Running, maintaining and scaling your own TensorFlow installation takes time and effort. Building and improving your own model is similarly difficult. There are questions of how you snapshot your model and roll it back if you mistakenly train it with the wrong data. Or how you push your model out to mobile devices if it needs to be close to endusers and then keep it updated there or better yet incorporate feedback from endusers into the model.

Here are some key questions you should consider as you figure out whether to build or buy when it comes to machine learning

1. Do you have the scale and is your operations team staffed to run this yourself? Can you attract top machine learning talent to your company?

2. Do you have more data than almost any other player in your industry for this particular problem or could you benefit from a vendor’s ability to train on lots of data across multiple customers?

3. Is this machine learning application unique to your business or something all your competitors need to do also? (e.g., fighting fraud, moderating forums) Will this allow you to competitively differentiate yourself?

And here are some questions you may want to ask of vendors you would consider

1. Is this their core business and do they provide high service levels (uptime, latency, security, etc.)?  How strong is their customer support?

2. Do they have a track record of improving model accuracy over time?

3. Are they trying to value price or is their pricing predictable and reasonable compared to cost, especially at scale?

You will want to take a hard look at these questions and use your engineering team’s views as just one input. For the time being, machine learning is an area in which the personal interests of your engineering team have a high likelihood of diverging from what might be best for your company.

Friday, September 22, 2017 - 11:35am

Today’s post is over at on “The Unknown Path to a Decentralized Future.” Here is the opening:

Some companies with currently centralized services have been criticized for issuing tokens and raising money in ICOs. There are even allegations that venture investors are pushing companies to do so as a ploy for liquidity. I suspect that some situations like that do actually exist, but I know from first hand conversations that many of the entrepreneurs pursuing this route are doing so out of a genuine conviction that it is the right path to a decentralized future.

Wednesday, September 20, 2017 - 11:30am

Last Uncertainty Wednesday I introduced how think about weather using the concepts we have developed in this series. We saw that our improved data collection and ability to process more complicated weather models has given us significantly improved predictions. We also learned that because of the chaotic nature of weather, despite our massive progress on short term forecasts we still do quite poorly for forecasts that go out further than a week.

Now a key question that we asked throughout this series is what we can learn about the reality of weather from the signals that we observe. Of particular interest here is the question whether the observations should us make more less inclined to believe in climate change, a topic that I have also covered extensively here on Continuations.

To get started on that we need to draw a distinction between weather and climate. The first sentence of the Wikipedia entry on climate provides an interesting approach here:

Climate is the statistics of weather over long periods of time.

The key words here is the “statistics” of weather. What is a statistic? It is a summary of observations, such as a minimum, or a maximum, or a total, or an average, or a variance.  

The weather is the temperature, rainfall, humidity, etc. on a given date and time. There is past weather, current weather and forecasts of future weather. The climate would then be summaries of these observations over longer periods of time. These periods don’t need to be contiguous. For instance, you could take the minimum, maximum and average temperatures for the month of July in New York City using data from July for say the last 10 years (or the last 100 years). 

So the weather are the raw observations and the climate are statistics computed from those raw observations? So the climate is simply a summary of the weather? That seems, well, not a very useful definition of climate.

The weakness of this definition is the result of a problem which I alluded to in my post about expected value, where I warned against confusing the sample mean with the expected value of the probability distribution.  A better definition of climate would be as follows:

Climate is the probability distribution of possible weather events.

The statistics of weather are supposed to help us understand what that probability distribution is.  With this definition, climate change becomes a shift in the probability distribution of weather events. And we begin to understand that inferring whether such a shift has indeed occurred is quite tricky. We have a bunch of hot years in a row. Is that just “dumb luck” (sort of like losing the coin toss at five matches in a row)? Or is it that the distribution has changed (the referee is tossing a biased coin)?  

Next Uncertainty Wednesday we will dig deeper into the relation between the sample mean and expected value using the context of weather and climate. 

Monday, September 18, 2017 - 11:35am

My book World After Capital has been an ongoing project for a couple of years now. I am excited that as of this weekend I feel it includes all the ideas I want in there. Some of them, such as strategies for overcoming the dominance of nation states, are so far in a protozoic stage, but at least they are there. Others, such as the chapter on the sufficiency of “Capital“, are in need of a substantial rewrite because they have fallen out of sync with other parts (see the rewritten chapter on “Needs”).

What is next? Rewriting the Capital chapter is high on my priority list. I am also now in need of a copy editor to go over the whole book from beginning to end. There are lots of inconsistencies in voice stemming from the iterative writing process. There are also ideas that could be expressed more clearly. And I also want to revisit some of the sequencing, although having explored many possible high level outlines I am pretty happy with the one I have now.

Finally, I have been getting more inquiries for printed and bound versions of the book. So I am starting to look into options for that. I am particularly interested in Print on Demand solutions, as it seems crazy to me in the year 2017 to print copies without knowing the demand. On the other hand though I love a book that’s printed on quality paper in an attractive font and well bound. Ideally one can have both and not have the print on demand feel cheap and fall apart quickly. 

If you know an amazing copy editor who is also fundamentally interested in the topics of World After Capital, please send them my way. If you have ideas for high quality Print on Demand, I am interested in that also.

PS If you know someone who is great with fonts, instead of a designed cover I want just text (and of course also for the book itself)

Thursday, September 14, 2017 - 7:35am

So far in Uncertainty Wednesday we have mostly built up concepts and ideas, with only one extended example.  Given the two massive hurricanes Harvey and Irma, the weather received a lot of attention, including the question to what extent the occurrence and/or severity of these storms lets us draw any conclusions about climate change. For that reason we will spend the next few Uncertainty Wednesdays looking at the weather using the ideas and concepts from the series. 

First, let’s put weather in the context of our framework, which consists of reality, explanations and observations. The reality in question is the complete state of the Earth’s atmosphere. The first thing to note here is that the atmosphere is not a closed system. It receives energy from outside (as a first approximation entirely from the sun, albeit much of that via heat radiated back by the Earth’s surface) and its mix of gases changes due to processes here on earth, such as photosynthesis and the burning of fossil fuels.

Second, let’s not that our set of observations about the state of the atmosphere and of the energy providing and gas changing processes is small relative to the scale of the system. This stands in sharp contrast to the many smaller examples used throughout the series where the system had only two states and we have a signal with two values. Here we have a system with a great many possible states and we have a lot less signal (e.g., measurement of say air pressure). It is also important to note though that how much signal we have available, has dramatically increased over time through technology, such as satellite imaging.

Third, weather is a classically deterministic system. Albeit one with explanations we do not fully understand. There as some explanations that are quite simple and well understood, such as air flowing from areas of high air pressure to low pressure causing wind. We also understand for instance how air above land and water heats up differentially also causing wind to form. But explanations for other phenomena, such as what goes into cloud formation and when clouds begin to rain, are less complete and less well understood.

Fourth, weather is a chaotic system as described in the introduction to the framework. As we saw there, small differences in observations will lead us to large differences in predictions about the future, especially the further out the future is.

Something that follows pretty much immediately from all of the above is: as we get more observations (and better computers for crunching them) our weather forecasts will get better. Here is a graph from an article that appeared in Nature which beautifully summarizes this:

We can see that forecasts have gotten much better between 1981and 2015 with a 3 day forecast (blue lines at top) going from about 80% accuracy to about 97.5% accuracy. We also see that accuracy drops off dramatically as the forecasts go further out and even though we se a big improvement in 10 day forecasts (grey lines at bottom), they are still pretty bad and have leveled out around 40% accuracy. Something else the chart shows is the convergence between Northern and Southern Hemisphere forecasts. Whereas the overall improvement is the sum of both better observations and better models, the convergence is largely the result of much better southern hemisphere data in the satellite age.

Monday, September 11, 2017 - 11:30am

Going to work today will feel odd because the weather reminds me so much of 2001. Gorgeous blue skies. Not quite summer anymore, but also not yet fall. The weather is one of my strong memories from 2001.

The other memory concerns later in the day. We knew a horrific act of terrorism had been carried out that killed thousands of people. And yet on the Upper West Side where we were living a the time it didn’t feel entirely real as we saw much of it on television. 

I am experiencing that sensation too this morning as I am looking at pictures of destruction brought about by hurricane Irma (following on the heels of images of biblical flooding from Houston and the leveling of Mexican towns by an earthquake). It all feels so extreme and surreal and yet it is the grim reality for millions of people.

So my thoughts are with all those who were affected by 9/11 and those whose lives are being upended right now.

Friday, September 8, 2017 - 11:35am

I have already written a bunch about ICOs here on Continuations, including earlier this week about the Chinese ban. Regulators are approaching this, not surprisingly, by focusing on their own country. While that’s understandable, it undermines one of the most promising aspects of using tokens as equity: a global capital market that’s readily accessible to everyone. 

If you are a wealthy investor or if you are a large corporation you already have access to a global capital market today. For example, a few years back I called my broker at Morgan Stanley and asked them to buy some Japanese robotics stocks for me that are only listed in Japan. Morgan Stanley took care of everything for me including figuring out how to buy Yen, place the trade, custody the securities and make the holdings show up in my account.  Or take Apple, a company that already sits on a huge cash pile. Nonetheless, the company has been issuing Euro, Yen and Canadian dollar denominated bonds for the last few years tapping into global capital markets.

If you are a smaller company though or an individual investor with a smaller account you tend to be restricted to your local capital market. In many countries those local markets are relatively illiquid and/or suffer from other problems. In the US, for example, we have made it difficult for companies to go public through both overregulation and through a broken IPO process. Combined with ongoing M&A activity, here in the US the number of publicly listed companies has be declining substantially.

Now the end goal probably shouldn’t be a global free for all capital market with scams everywhere you look and people losing their life savings overnight. But conversely I don’t think the right answer is trying to stuff everything back into country-by-country securities regulation, much of which dates back to pre-Internet days and doesn’t at all account for how much more data can be shared today by a company on an ongoing basis. For instance, the bulk of US securities regulation dates back to the 1930s and the aftermath of the stock market crash. 

So directionally what could be done? As a starting point companies themselves could embrace high standards of transparency and could limit how much exposure individual investors can buy. The latter could either a fixed number or be based on suitability tests (eg if you can prove you hold a lot of say BTC or ETH you can invest more). As a next step there could be a global self regulatory body. People sometimes rightly scoff at the notion of selfregulation but there are great examples of where this has worked well and better than government standards, such as Demeter (for organic food). There doesn’t need to be a single one of these but there can be multiple which allows for some degree of experimentation.

One important question that comes up is how rights would be enforced in such a world. As a US shareholder in a US corporation, I have certain statutory rights. In practice though, the primary mechanism in the capital market is not voice but exit. If I don’t like what a company is doing my general recourse is to sell their shares. That mechanism also exists in the token world. The right approach to voice would be through the selfregulatory bodies I mentioned above, which could then be supported by traditional government regulators.

All said then, I believe that there is an opportunity here for a new system to emerge and it would be a shame if we shoved it back into the existing boxes before we have given that a thorough shot.

Wednesday, September 6, 2017 - 11:30am

Since I am an immigrant to the United States I want to comment on the Trump administration’s announcement yesterday to end DACA  (Uncertainty Wednesday will resume next week). I first came to the States as an exchange student in 1983 to 1984, when I lived with a wonderful family in Rochester Minnesota. One of my lasting memories is how quickly one can fall in love with a country as a child or young adult. For the DREAMERs, who all came to the United States as children and have grown up here, the United States is their home. Threatening to take that away from them is cruel. Doubly so after giving them hope first.

The blame for this situation though rests with Congress and past Presidents who have failed to make any meaningful progress on immigration reform. Right now, it is worth remembering now that the DREAM act has been around for 16 years. There have been multiple attempts to pass it with at varying times support in the House and Senate, but never the two at the same time, including a bipartisan filibuster in the House that included 8 Democrats. The opposition by Democrats often arose because they wanted comprehensive immigration reform or nothing.

There is much more we need to fix about immigration, but passing the DREAM Act would be a good start. Everyone should be calling their representatives and senators and urge them to do this. It has broad public support. The key challenge will be not having it become a political football again with all sorts of crazy plans attached to it such as funding the border wall. That’s where pressure from citizens on their representatives will make a big difference, as well as senior leadership from both parties.

Let’s get the DREAM Act passed and let it be the beginning of a return to bipartisan politics!

Tuesday, September 5, 2017 - 11:35am

As has been widely reported, China has banned all ICOs. Given the torrid pace of ICOs in China, many of which appear to be downright scams, this should not come as a surprise. What will be interesting to see though is what comes next. Here are some key questions.

First, is this a temporary ban only? Will China come out with a set of regulations that allow some ICOs to go forward? Given that China has allowed crypto currencies per se, I expect that they will come up with regulation and that this ban was them pulling the emergency break.

Second, how will regulators in other countries react? They might see this as an opportunity to follow suit, or they might see it as an opportunity to position themselves as more attractive to innovation. Given the cautious initial findings of the SEC around applying the Howey Test, I think they are unlikely to act rashly now.

Third, regulation will be a long term issue for innovators and investors in the crypto currency field. Because these systems are global from day 1, countries will find that they have less regulatory power than in more traditional financial markets. So I expect that what’s legal where will shift around a bunch over time. That adds an important component of risk and uncertainty to innovating and investing in the blockchain space. How big is that extra risk and what to do about it? That’s one of the key issues to wrestle with.

As we continue to invest in this space we will be paying close attention to this. So expect updates here in the coming months. 

Friday, August 18, 2017 - 11:35am

This will be my last post until Labor Day. I will be spending as little time online as possible, reading books instead, spending time with family and friends and working on World After Capital.  I will be disabling all notifications on my phone and checking email only twice a day. Given all the craziness here in the US and elsewhere in the world, I have been spending too much time on news and I am looking forward to this break to dial things down. 

Thursday, August 17, 2017 - 7:35am

Last Uncertainty Wednesday, we saw how diminishing marginal utility of wealth provides an explanation of risk aversion via Jensen’s inequality. Why would it be then that lots of people seem to like small gambles, like a game of poker among friends. One possible explanation is that the utility function is locally convex around your current endowment. So this would look something like the following:

In the immediate area around the endowment (marked with dotted lines for two different levels) the utility function is convex, but for larger movements it is concave. 

In the convex area someone would be risk seeking. Why? Well because Jensen’s inequality now gives us

U[EV(w)] ≤  EV[U(w)]

Again, the left hand side is the utility of the expected value of the wealth, whereas the right hand side is the expected utility, meaning the expected value of the utility. Now the inequality says that someone would prefer an uncertain amount over a certain one. Here is a nice illustration from Wikipedia:


We see clearly that the Certainty Equivalent (CE) is now larger than the expected value of wealth, meaning the Risk Premium (RP) works the other way: in order to make a risk seeker as well off as accepting the bet, you have to pay them more than the expected value.

Next Uncertainty Wednesday we will look more at how incredibly powerful convexity is in the face of uncertainty. 

Albert Wenger is a partner at Union Square Ventures (USV), a New York-based early stage VC firm focused on investing in disruptive networks. USV portfolio companies include: Twitter, Tumblr, Foursquare, Etsy, Kickstarter and Shapeways. Before joining USV, Albert was the president of through the company’s sale to Yahoo. He previously founded or co-founded five companies, including a management consulting firm (in Germany), a hosted data analytics company, a technology subsidiary for Telebanc (now E*Tradebank), an early stage investment firm, and most recently (with his wife), DailyLit, a service for reading books by email or RSS.