Transparent Big Companies, Or Lack Thereof

Transparency within companies is a hot topic. Many popular startups tout a transparent company culture, for instance in an interview with Keith Rabois:

In the early days of Square, transparency across the organization was taken for granted. It just seemed obvious to both Dorsey and Rabois that is how you build a high-functioning company — you make it incredibly transparent. Ultimately, if you want people to make smart decisions, they need context and all available information. And certainly if you want people to make the same decisions that you would make, but in a more scalable way, you have to give them the same information you have. Complete information also helps reduce the politics in an organization. One of the key drivers of politics in an organization is information asymmetry.

Why then does this behavior not seem to carry over to large companies? Possible explanations that come to mind:

  • More people mean it is harder to find the right person to talk to/get information from, which prevents effective info sharing.
  • More noise, less signal. Just like the general growth of bureaucracy happens over time, so too does the growth in non-essential/useful information, which discourages seeking out new information (as it is rarely useful to you), which creates a low transparency impression.
  • More people means fiercer competition for status and recognition, which discourages info sharing as others are worried it could be used against them/weaken their position.

Of course another explanation is that big companies aren’t necessarily not transparent. Google and Facebook both have a reputation for cultures that strongly value information sharing. This seems to go hand in hand with strong culture and norms of behavior, which is fundamentally unsatisfying as an answer, but might be the best we can get to right now.


Transparent Big Companies, Or Lack Thereof

Monte Carlo Python

To calculate a Monte Carlo simulation in Python:

  1. Get your Confidence Interval for the variables.
  2. Use the sci ppf (percent point function), with (rand(), UpperBound + LowerBound / 2, numpy.std())
  3. Calculate 5000 (or more) potential outcomes using a python iterator.
  4. Repeat for all the relevant variables.
  5. Aggregate and group the into reasonable increments, and create a cool histogram.


Quick and dirty way to use a really excellent analytical method!


Monte Carlo Python


I recently learned about a nonprofit Startup called DemocracyOS that is attempting to improve the democratic decision making process through an open source platform for sharing and voting on information. DemocracyOS seeks to improve democracy by tapping into the wisdom of the crowds through an online platform that allows markup of bills and voting on policy proposals. It also seeks to stimulate debate between participants, and form new groups of motivated civic participants.

I really like the DemocracyOS interface. UX is an under appreciated aspect of useful decision making structures. Creating low barriers to entry is the only way you’ll get large scale participation, and when you’re not getting many direct benefits out of it, for instance with voting, even a little friction will turn people away. The DemocracyOS team seems to appreciate this, and has created a slick modern experience that makes it easy to participate. For instance the Buenos Aires City Congress site has a number of interesting proposals on it that are easy to read (well, with Google Translate working) and the comment section has the benefits of a Reddit style up-voting to avoid the cesspool of general, unweighted comments.

The team behind DemocracyOS seems to hold several assumptions  about democracy. The first is that more direct democracy is a good thing. The second is that if people had more information, they would make better political decisions. There is a long history of debate around the first issue, which doesn’t need to be repeated here. And as for the matter of information flows, I worry that information alone rarely changes opinions. Even if DemocracyOS spreads better information and helps more people coordinate, I’m skeptical they will actually change their minds.

DemocracyOS is an interesting project, and I think it is trying to tackle an important issue in improving social structures. My hope is that this will be more than a ‘reddit/rapgenius for policy’. But even if it doesn’t it’s github page is excellent, and they’ve done a bunch of excellent work on useful open source projects. So they’re cool in my book.


What is Futarchy?

“Vote on values, bet on beliefs”.

Futarchy is a hypothetical form of government proposed by Robin Hanson that uses Prediction Markets to coordinate government policy.

One of the problems with government is it is difficult to determine whether a certain policy actually made a difference. We can speculate whether tax cuts helped or hurt, but there are a million different reasons why an economy might have improved. It’s is too hard for even experts to tell, much less the public.

The solution is to use prediction markets to aggregate information, and then on the basis of the result implement specific policies. Voters would vote on common metrics, for instance a national GDP, and then speculative markets would be set up to bet on whether policy X or policy Y would improve that metric. If the policy hit a certain threshold of support it would be implemented. Then, if it actually did make a positive improvement – maybe over a year timeline or any arbitrary timeframe – the accurate bettors in that market would be rewarded.

In this way we could use the market mechanism to effectively aggregate disparate information about national policies, coming to better decisions than just relying on a few experts. This is a way of leveraging that strength and applying it to policy making.

What is Futarchy?

Ramen Noodle Run Rate

How much money do you need to survive? Thats an idea on my mind recently, as a friend left his job to find himself. Am I jealous? Maybe a little, definitely a romantic idea.

Rent: Lets say you’re living in a shitty part of a big city and you’re sharing a room. I think $900/mo – 1100/mo, including utilities, seems reasonable.

Health Insurance: Looking at the individual health care exchanges, it seems that 100/mo – 300/mo is a reasonable estimate.

Food: Soylent would be approximatly $162 a month to live exclusively on, and guessing that you might want some non-powder food I’ll estimate $200/mo – $300 /mo

Transportation: You need to leave your room occasionally. 100/mo – 200/mo

Sanity preserving miscellaneous expenses: This includes all odd random expenses, like movies you didn’t sneak into. 100/mo – 200/mo

Using I modeled, with monte carlo simulations, the likely expenses.



Seems surprisingly reasonable!

Ramen Noodle Run Rate

Futarchy Metrics

In a ‘futarchy’ system voting on national metrics is how we would determine what the individual policies should maximize. That means that the voting populace would have a variety of different types of metrics to select from, likely grouped into different categories based on the type of policy being evaluated.

So for instance, if an idea to raise the property tax in your neighborhood is being discussed, premised on the idea that it will raise a certain amount of money for schools, then there are a number of metrics that could be implemented to evaluate the effectiveness of the policy. For instance if the primary question is how much money would be raised then a straightforward metric would be “Will Policy X raise 10 million over five years”? However if the question is how much will implementing this policy improve educational outcomes, then a metric such as “Will Policy X improve SAT scores by 10 points” could work.

How do we separate the noise from the sound? High fidelity markets, combining multiple metrics and parameters, could work. If there are multiple metrics, than you could have conditional markets where only if one metric was first achieved (“It does raise 10 million”) then the second market would come online (“Improve SAT scores”).

Futarchy Metrics