I’d like to talk to you about a project I’ve just started with my partner, Toby Segaran, called Freerisk. The project is desgned to explore ideas on how technology and web centric ideals can fix what we feel is the biggest problems in the financial world today, the failure of credit rating agencies to protect us from the very risk investments that created the current financial crises.
This is something we’re interested in because (... the financial world is in chaos right now)
We feel this is an exciting time to explore financial technology: it’s destroying itself; an d it’s revealed it’s fundemental assumptions of secrecy and opacity to be flawed. No sector of finance shows this more intensely then tha bond rating industry. This is an industry so bad that Rep. Henry Waxman, chairman of the U.S. House Oversight and Government Affairs Committee to say: \"The story of the credit rating agencies is a story of colossal failure,\" It’s the industry that gave AIG and Lehman brothers top grade credit ratings up until the day before Lehman went under. But we choose to see this is a good thing. And now we have the opportunity to recreate our finance industry from the bottom up. And we have a choice: a path of openness and information sharing, or more opacity and secrecy. We choose openness. We’ve kicked around some ideas on how to solve this and what we’ve come up with is:
So we propose a simple idea that we hope can be part of a solution. We started building a simple platform that combines authoritative data, user contributed data, contributed algorithms and a test framework that firmly places risk metrics into a new arena of financial commons.
Although I’ll be showing a couple of screenshots, I want to emphasize that it’s just a prototype, but I hope it’s enough so that you see we’re really trying to build something and not just get angry.
Our goal is to reinvent a financial system where... (risk metrics are part of a financial commons)
Risk should be in the commons
The information ratings agencies produce is more valuable to society as a whole than individuals. The incentives to payment don’t adequately compensate society for the harms of privatization.
This is why commons exists.
Fundamentally: “Risk Management is too Important to Society to be a Competitive Advantage”
So for the rest of the talk, what I’d like to do is walk through what makes the credit raters so important; how they function in a structurally flawed market, and then address how freerisk solves some of these problems and where we can go from there.
So here’s what credit rating agencies should be doing.
Moody’s, S&P and Fitch sit between the world’s largest pool of investment money, on the right, and the worlds largest need for funds on the left.
Government investing, pension funds, investment accounts: these are invested by money managers on behalf of those who don’t know how to tell a good bond from a bad bond. so to keep tabs on what the investors are doing we’ve evolved a system where credit rating agencies rate investments on how risky they are, and investors agree to abide to a given level of risk. this way we can all rest assured that our money isn’t being put at risk.
So if you invest in mutual funds, govt. bonds, or your retirement account does, you’ve bought into their logic. Either by law of covenant, this funds may only invest in highly rated bonds.
And if you, or your company needs debt to reach a goal, you need to fulfill their criteria, or you will be cut off from the investment markets, drastically increasing your interest rates.
Now don’t be fooled by the fact that we talk more about stock prices than bond prices. Bond markets are way bigger, at $45Trillion world-wide. That’s $45Trillion under the control of 3 firms.
That’s led the the following realization... (Moody’s is a super power)
Here’s a quote from 1996 pointing out that in the post-Soviet, globalized era, the only thing more important than credit is military power. perhaps.
And the market has tripled since then.
The result has been that the regulating force on our new, interconnected economy has become a private enterprise super-power. Revenues have doubled over the last 5 years alone.
Now, if the system were perfectly balanced and well-structured, this could work, but as I said early, the system has some flaws, and these flaws create horrible problems:
Adn the system has grown organically, and without adopting to modern discoveries and techiques, creating a structurally flawed system.
(It’s important to realize that there are a lot of reasons to have moral outrage, but that a lot of these problems are structural. You have every right to hate the credit raters. But you should be angry at the system.)
The credit rating system was designed to create incentive problems.
There are a number of ways you can slice the data, but we’re going to show you what we believe are the 4 most important structural problems in our financial system.
To explain what I mean by this, let’s imagine that I’m a company looking to get a bond of mortgages I have rated.
Regulators want the rating data to be free, and raters want them to be free so they can’t be sued for giving a bad rating
So for Moody’s to afford to create a rating, I have to pay to get my bond rated, because consumers of the data shouldn’t pay, or the market will be impaired.
But since I’m paying, and there’s more than one credit rating agency, each will compete to provide me with a better product, and in this case, a better product is a better rating. Which of these raters would you use?
Rating agencies understand this, and make their terms favorable to bond issuers.
So this looks and smells like a bribe. But it’s perfectly legal.
Rep. Stephen Lynch described a \"shopping around\" scenario, in which a firm seeking a rating takes its business away from a credit agency if there is a chance it will not be granted a good rating. This influences the credit agencies to give high ratings or risk losing business, lawmakers said.
And before you get too angry at the rating agencies, you should realize that these guys are just responding to the market forces presented to them. Which leads us too our first structural problem...
“Payment systems create bad ratings”
It’s very difficult to design a compensation system that doesn’t create incentive issues. Almost any commercial system will create game-able opportunities, and there’s too much money flowing through this system to tolerate that.
The only acceptable solution is to avoid any explicit payment system.
Now we wish that were the only problem, but we have to go on...
The rating agencies have a monopoly - and they have to be used by large pension funds, and most mutual funds. And regulation makes it hard to create a competing ratig service. You can’t get away from them.
But the quid quo pro of goverment created monopolies doesn’t hold here.
These rating agencies have no legal obligation to describe their methods.
The patent system in the United States protects a company’s right to use new inventions, but it also REQUIRES them to say exactly how the invention works. Yet, strangely, we don’t require this for rating agencies. We spent years hearing about how terrible patents are, and yet here’s a system that has all the bad parts and none of the good parts.
So we’re stuck with being told “this is triple A. Buy it because we said so.”
Here’s an example of that, it’s Moody’s statement abotu Lehman Brothers the day after it went bankrupt.
There’s almost nothing to say about this... it makes me extraordinarily angry... What this means is...
Moody’s was basing it’s high rating on a belief that the government would bail out the bank. Which is fine, but investors didn’t know that, and didn’t know what sort of risk they were taking on.
This sort of information needs to be made public, it needs to be clear - so that we know what KINDS of risk we’re dealing with, instead of just how MUCH risk.
So the second structural problem is this: there is no transparency.
I’m not allowed to know what your method is When you give me ratings I have no idea what they mean
This isn’t an acceptable situation - we’ve learned from open source, and commons data the value of transparency, and we need to take those teachings here:
The only acceptable solution is full disclosure of rating methods
But what we do know about their methods isn’t much prettier...
Because, at least where it counted in this financial crises, they were all using the same equation. The Gaussian Copula model.
$12Trillion in mortgages, all evaluated with one equation. It’s staggering, even if the equation worked. But the equation turned out to be wrong.
Now, there were good reasons to use this model: The Gaussian copula was used everywhere because it was easy to calculate and easy to understand, and those were the pressures that fueled idea dissemination in the financial industry over the last few years.
But because it WAS wrong, and it was used exclusively, it REQUIRED the creation of an asset bubble...
See, the world we live in only lets most investors invest in a only certain class of bonds. But these bonds are, as a result, far more expensively than other bonds in the market.
Now if everyone rates everything the same, and there’s no extra information in the market, then this situation is fragile, but stable. But...
But what if there’s a bond that mistakenly looks as safe and is rated as safe as the safest bonds, but as cheap as the bonds you couldn’t invest in before? You would want to invest in it. You would have to invest in it.
And because there’s only one source of information, and all of your competitors share that source of information, and have the same constraints, they would have to invest in it too.
Pretty soon, it would be the only type of bond anyone would want to buy. We have a name for this phenomenon, it’s an asset bubble.
Which then leads us to our third finding, “You need ecosystems to have a stable market.”
The paucity of ideas doesn’t allow you to have diverse ideas, and the only alternative is destructive group-think.
We need intellectual redundancy in a system this significant... because even if we get everything right...
Risks are hidden. Even If you nail down the right algorithms, analysis and interpretations you still need the right data to get your risk.
Now, someone I consider brilliant, Nasim Taleb, will tell you that we simple CAN’T know the data we need to evaluate risk. And there’s a lot of validity to this. But it’s nihilistic to simply given up on predicting. We can take an alternate take.
As a responsible investor, you might want to research some of the risks on your own. You’d look through long financial statements, try to track down what something in one footnote referred to in another document in order to find out that a class action lawsuit was being brought against a company or something like that, but in the end you might miss it.
Obviously everyone missed a lot of them, this time around. But some people got it right, and I don’t want to belabor the point, but I just want to quickly tell you the story of John Paulson.
John Paulsen is a hedge fund manager - Michael Lewis wrote a long article about him Portfolio magazine that some of you may have read.
In 2005, he figured out that people were buying a lot of highly-rated credit-default swaps that were clearly terrible investments, paying way too much money for them.
So he did what any savvy investor would do upon discovering that something is way too expensive. He started selling those credit-default swaps and made his hedge-fund the largest single-year hedge-fund return EVER.
That’s great for him, but other’s needed acess to this data too. We need ways to surface this information to others in a coherent way. We need a way to reflect biases and hunches in our models of risk. But ratings agencies don’t give you a way to incorporate your own data into their ratings.
So our 4th and final reason that ratings are bad is because the information used to create them is not diverse enough.
Sources of information are unexpected. We have a lot of people reading, researching and thinking about the activities of companies. But as far as we know none of this is ever used as part of a model.
The only solution here is to allow contributions of data from many sources
Ok, so it’s completely broken, and inevitably leads to massive problems with our financial system.
That doesn’t mean there’s time to cry about this. We need credit to work. We can’t function as a society without trust.
The only real question that should be facing us is what exactly are we going to do about this?
You would be forgiven to run away from these problems.
“Payments are a problem” “Consensus is a problem” “Data Sources are a problem” “Opacity is a problem” “And by the way it was to be accurate”
But that’s a pessimistic view of the word. Engineers, designs; people who construct things, we don’t have problems, we have requirements. Let’s look at this another way:
Make a system that’s:
Accessible Open Supports Diversity Transparent And is Accurate
These are requirements that we can work with. And thanks to ideas developed in web technology and community building we know how to implement a lot of these.
So let’s see how we did in addresses these requirements.
Commons allow free creation, re-mixing and free collection of data. And open data makes the fundemental data available to everyone.
We can shine transparency onto the role of subjectivity and bias by building a system that accepts user-data but marks it as such, and by allowing ANYONE to create a credit rating we can assure that there will be some diversity in our ratings. And if we can test them against the data, then the scores will be diverse enough to support everyone’s investment bias.
So what does this look like:
Formost. Freerisk is a story about opeb data.
Open data means both consuming the data that’s made available on the web and republishing so it’s easy for others to use.
The first step to Freerisk was getting authoritative financial data, so we looked at what the govt is currently providing.
All public companies must file with the SEC their annual and quarterly reports, special events, and when insiders buy or sell shares.
As of today, these are mostly hard-to-parse text-files, but the SEC is mandating that all companies switch to a format called XBRL. In 2009, all the biggest companies will be required to switch to XBRL filings and in 2011 every public company will have to. XBRL is a machine-readable XML based format with its own set of namespaces for accounting terms. You can even pull them using an RSS feed. But...
This is an XBRL file. It’s a quarterly filing for 3M -- they make scotch tape.
It’s 1070 lines long. A cottage industry now exists in producing and parsing XBRL so companies can meet the SEC mandate.
Further, XBRL isn’t just long, it’s also extremely complicated... have a look at those numbers.
This won’t work for creating the next generation of financial hackers.
The effort in tracking down a taxonomy and trying to match it to the one you’re using is enough to make you cry. I know this from personal experience, but we’re doing it for you anyway.
So we started simplifying.
We took a semantic data-store and started filling it with the basics. The things that appear on an income statement or a balance sheet, in a much flatter, easier to follow graph-like structure.
And because it’s a semantic store, it will easily accommodate new kinds of data, and also allow users of the system to create their own assertions about things they find hidden in the footnotes
We tried to use standards wherever possible.
The data itself is stored in RDF, we used the Generally-Accepted-Accounting-Practices (GAAP) namespace for accounting terms where possible, and the Freebase namespace for companies and organizations.
This means the data can always be queried with SPARQL, even as it gets more complex.
Here you see a query for companies with a low current-ratio, which is the ratio of their current assets to current liabilities. We can construct very sophisticated queries across the whole dataset.
We also tried to create simpler, more familiar ways for developers to access the data.
Here’s the JSON result for the 3M statement, through an API we build.
You can see we’re still using the GAAP namespace, along with numbers for values. This is much easier to read and deal with for someone inexperienced with XBRL.
All this data then let’s us build something new and special. Because with an api we bring allow anyone into our system to rate and learn about corporate risk using entirely open APIs built around restful end-points.
Our community isn’t necessarily composed of programmers though, so we’ve taken steps to make things easier. Our APIs will require minimal effort on the developers part and provide unfettered access to all available data freely. Using these calls will provide you with risk scores and financial data.
But we allow you to add to the commons data by implementing your own endpoints, which Freerisk calls on your behalf.
A risk calculator is simply a set of two urls.
One returns the query to pull the data you need to perform the risk calculation. The second url let’s us postback the data to your calculator. Your calculator then replies with the risk score.
And You don’t need to know how to write your own server software. This isn’t rocket science.
We provide templates that you can use to implement your risk strategies in.
Here’s one example in Ruby on Rails.
It’s just as easy to see anyone else’s results. We don’t even have to know the calculator exists.
Just pass the url of a calculator to us, along with the time frame and the company name, and we’ll return to you the risk scores on behalf of the calculator.
In exchange for being intermediated, we take care of the data management and integrity issues for the calculators.
And we’ll tell you how well you’re doing.
If you can make your algorithm obey a few simple behavior patterns we can provide more.
We’ll provide a metric for how well your score predicts credit failures provided that your rating
Is simple and monotonic (goes up only)
and your score goes up if a company is stabler.
And if you can give us a default probability, we’ll go ahead tell you how correct you are.
You don’t need to be a math whiz to write a great model.
This model was the first model we implemented. It’s a bunch of logically derived if-then statements.
And any approach is fine. This model is a simple regression score resulting from discriminate analysis more commonly used in computer science than finance.
was first written in 1968.
Five variables, a simple score. This one can even give you sensitivity to certain parameters, which is a huge advance over ratings.
And it works...
There’s the Lehman brothers Z-score over the last year before it failed.
The red line denotes almost certain failure. The green line here denotes almost certainly viable.
Lehman isn’t even close for a year.
Right now what we have is a lot of anger, a simple idea, and the beginnings of a prototype...
But we know what needs to be done to make this better.
The biggest hurdle right now is data completeness.
The more data we can collect, the compelling this project is -- even beyond its use for credit ratings
We would like to eventually support data from many sources, government and private, from both inside and outside the United States
We need more algorithms too.
We’re speaking with about one algorithm developer a week to get their ideas online, and we’re comfortable with that pace.
We don’t care how biased or opinionated your idea is.
(optional) My favorite conversation we’ve had about Freerisk was with a guy who’s incredibly concerned about deflation and credit risk. Now you may or may not agree with this. But this guy deserves a change to see if he can build a new model to find safe bonds in a deflationary world, and Moody’s won’t help him.
We’re his only shot.
We’ve shown that even simple calculations t can be competitive with the official guys.
We’ve shown that there are many people out there who find the risks and publicize them long before they’re officially recognized.
We think there’s something here. And we think it’s worth a shot.
I hope you’ve found the talk interesting, and I encourage you to build something just as ambitious and crazy, because this is the time to do it.
I Just Don’t Trust You:
Can Engineers Change Credit
Follow us on Twitter:
#6 @jandersen, @kiwitobes
“The story of the credit
rating agencies is a
+ A commons approach to credit
+ An open ﬁnancial-data source
+ A user-contributed ﬁnancial data
+ A set of contributed rating
Standard & Poor’s
There are two superpowers in
the world today in my opinion.
There’s the United States and
there’s Moody’s Bond Rating
Service... And believe me, it’s
not clear sometimes who’s
–Thomas Friedman, 1996
NRSRAs Enjoy Patent-Like
Without Patent-Like Disclosure
Today’s rating action follows the
collapse in market conﬁdence in
the ﬁrm, and Lehman’s
announcement that it was ﬁling
for Chapter 11 bankruptcy
protection after its failure to
reach a merger agreement with a
stronger strategic partner.
–Moody’s Press Release, 9.15.2008
IN OTHER WORDS:
In fact, a miracle did not
FRS: Rating Creation Needs to Be
FRS: Rating Data Needs to Be Open
FRS: Rating Scores Need to Be
FRS: Rating Processes Need
Data Needs To Be
Open Open Data Source
Creation Needs To Be UGC Data Source
Scores Need To Be Algorithms
Scores Need Accuracy
You too, can make a risk
def query(company, period)
render :text = sqarquery
Check out the work of others
Your score goes up
with defaults rates
for better companies
Your score is
measured of defaults rates
1. Positive Net Income (+1 for each)
2. Positive Cash Flow
3. Return on Assets up
4. Cash Flow Net Income
5. Debt / Assets up
6. Current Ratio increased
7. #Shares Outstanding same
8. Gross Margin last year
9. % increase Sales the % increase
EBIT / Total Assets x
+ Net Sales / Total Assets x 0.99
+ Value of Equity / Total
Liabilities x 0.6
Microsoft AIG Lehman
Nov 31, 07 Feb 28, 08 May 31, August 31, 08
(Year Before Lehman Collapse)