ZDNet has just had a bit of a coup with a fascinating article, white paperĆ and webcast, explaining how Goldman Sachs calculate and manage risk.
The webcast finishes with a view that by being involved in selling Credit Default Swaps (CDS) on the one hand whilst moving collateral risks off balance sheet through CDS on the other, they managed to dump the risks before the boiling pot melted.
Hereās a summary:
Goldmanās success dates back to the mid-1940ās, and is based on their recruiting policy which is far tougher and more rigorous than any other firm has been over time. Other firms may claim they hire top quartile or decile, but Goldman go for the top 5% and then sort them out over 2-3 years once on board. Thatās why they have had so many star players, such as Lloyd Blankfein, David Viniar, E. Gerald Corrigan and Hank Paulson.
They are also very disciplined. People at that level are going to normally have an ego, but no egoās are allowed at Goldman.Ć The word āIā is not allowed, itās always āweā. Theyāre not even allowed to see the eye doctor.
Itās all about teamwork too. They have all kinds of divisions, with two or three co-heads running each area simultaneously. Itās hard to make that work but, when it does, it works beautifully.
Thereās no hierarchy, and everyone is expected to take leadership in decisions that matter the most. The intensity with which they communicate all the time, 24*7, is amazing too. I mean everyone spends half an hour before sleeping and immediately upon waking doing voice mails and emails as their priority.Ć That's dedication.
How do you take this culture and communication capability, and then apply data systems to underpin it?
Well, flight simulators donāt make you a pilot and models donāt make you a financier. Thatās why Goldman tempers the financial models with business judgement, and the feedback models with cultural overlay. Those are two different models and both have strong human input to them.
Thatās why, according to the latest figures from Bloomberg, Citi has taken $84.4 billion in write-downs, Bank of America $40.2 billion, JPMorgan $29.5 Billion, Morgan Stanley $20.5 billion and $7.1 at Goldman so far.
Goldman Net Revenue from Trading:
2007:$31.2 billion
2008:$9.1 billion
4Q 2008: -$4.5 billion.
They also have a very strong capital commitments committee that is actually the most important decision making committee in Goldman.Ć You may think itās the Board, but itās this committee as they are responsible for keeping Goldman's capital afloat.
By way of example, on the day the markets dived 23% in a single day, every part of that negativity had been scoped out in a report that day and sent to the committee showing what could go right and what might work. Thatās a very different mindset to thinking about what could go wrong.
If you go to most trading, the reporting of most risk management is through the head of trading or the head of that business unit.Ć That creates a very inefficient system as a lower status individual ends up arguing with a higher status one, and you can guess how that works out.
Risk management needs to report directly to the top ā the CEO or CFO ā so that risk communication can happen without getting overridden by alpha male characters. Thatās the criticality of why Goldmanās team culture works.
From a technology perspective, Goldmanās key risk technology is based upon SecDB, a database and pricing system originally rolled out in 1996 which is now used across the firm in all their security trading businesses. They then use GridServer 4.0 to balance the load for risk calculations, as well as a whole range of relational databases developed in the 1990ās, to keep track of every product they sell to whom.
In the context of risk, they use a variety of methods to assess risks including historical events, predictable events, monetary values and various economic parameters. These are used for stress testing purposes against each instrument and trade.
The way you stress test is to shock an asset class with an event that has occurred in the past or that you might see happening in the future. An example might be a critical change to the dollar:yen. You throw that volatility against an asset class portfolio to see how that shock might trickle down through the portfolio, and the direct and indirect impacts.
Equally, for monetary values, you might shock a particular stock and see how that affects all of their stocks and bonds, and any derivatives and related instruments against that stock.
Finally, economic parameters would includes things such as a rally for the euro, gold pricing dipping and such like.
The thing is that everyone can get the models, but when you have the fastest car you need to bet on the driver, not the car and, in the case of the Lehman collapse, no driver had thought about the risks of who owned the collateral wrapped up in multiple Credit Default Swaps (CDS).
How do you get down and take collateral credit?
These were fields were not carried over between different instruments, which is why it was all buried below the line.Ć It also showed that no matter how good your model, sometimes you cannot build in an unanticipated risk impact. For example, you put in details of the previous CDO that a trade related to, but that information would not be carried over to the next trade.
Risk management also has two main components and often people look at one and not both. The two elements are the economics and the quants related to each instrument and trade.
Thereās a fundamental economic analysis to see whatās happening in a portfolio, and then a quantitative approach to risk management that should be independent of those fundamentals. You cannot do effective risk management when focusing on only one of those.
Thatās the reason why some of these models have failed. If you only look at one of these on their own in your risk framework, and just stress test one side of the equation with key drivers such as a house market failure, then you fail.
How Goldmanās work is that they:
They then stress test these portfolios and investment and their ripple effects based upon industry, market, credit and business risks, and see the impact upon default rates, ROI, recovery rates and more.
So, for example, they would continually put in stress factors such as: what would happen if there were a major rise in default rates? What would our strategy be?
They would know the sensitivity of a whole portfolio for rises in credit defaults and the actions to take, along with the likelihood of price movement in the portfolio.
Thatās why itās critical to stress the portfolio based upon high default rates, even if you cannot see it happening, in order to know the ripple effects and price movements on other parts of the portfolio.
Some stress test by dropping prices or raising Value-at-Risk VAR) and doubling volatility, but you need to do all these together because price movements and volatility are independent of each other.Ć You need to know the ripple effects each have on the portfolio and that's what Goldman's do.
But if these systems are all the same, why would Goldmanās build their own rather than using a package?
Itās not that systems donāt matter. Of course you want the fastest car to win the race, but everyone will be pushing towards the upper quartile of performance. Therefore, what distinguishes you in the race is how you drive that car, not just having the fastest car.
Everyone will want the best systems, but it has to be the best systems that fit your culture. Thatās why you have to overlay human judgement, and allow traders to say that if the market is doing X and be doing Y, then give them the latitude to deal that way.Ć That's what Goldman builds into their systems, based upon a team culture.
Itās not only using the best models but knowing what the competition will do and being able to make judgement calls on that.
How will the other players react and can you anticipate that and get to the door first?
Most risk managers focus upon the details in the model but the top risk managers, such as Goldman's, focus upon the assumptions in the model. What are you assuming will hold true when you look at the outcomes of a decision and relying on that model?
The model will only work its outputs based upon those assumptions and this is why that is key, especially as the model works on correlations of risk that were input at a particular moment in time. Those correlations can change over time and you need to be aware of that, and change them accordingly.
There are a lot of components under assumptions that are direct drivers into the vulnerability of your risk structures.
Thatās why Goldman is winning out, because if you look at the Big Five investment banks ā Bear, Lehman, Merrill, Morgan Stanley and Goldman - and how this has played out, Goldman is the stand out (comparative results for third quarter 2007):
Itās not to say they didnāt take a hit so if you cannot see the rabbit, look for the footprints in the snow.
Now we know Goldman are good but they may have taken some excess risks albeit by accident. For example, commercial banks talk about total accumulation to counterparties. Investment banks just talk about a trade and, as long as both sides offset, then youāre done.
Goldman over accumulated risks with AIG through CDS as a result. This footprint is shown by the fact that Goldman were far less leveraged than their competitors but still pricing competitively, so they must have had some off-balance sheet cover through CDS.
This would make up for the implicit versus explicit imbalance between prices.
Now, AIG was good for the counterparty risks because that meant Goldman did not have to put up collateral.
What probably then happened, when Lehmans collapsed, is that another part of AIG had collateral calls which triggered the domino effect.
Goldman Sachs, being in both the collaterals side and the CDS side, could very rapidly use their culture of fast communication across the internal human network to rapidly understand how those domino's were falling. They could see the dominoās start to fall and how, in a few days, it would hit elsewhere.
That gave them the critical jump to get out of the door fast before it hit their house.
Chris Skinner
www.thefinanser.co.uk
The webcast finishes with a view that by being involved in selling Credit Default Swaps (CDS) on the one hand whilst moving collateral risks off balance sheet through CDS on the other, they managed to dump the risks before the boiling pot melted.
Hereās a summary:
āIn the third quarter of 2007, Goldman, by sensing in its bones that a collapse was possible, earned a $1 billion profit, by betting against mortgage-backed securities. Merrill Lynch took a $2.2 billion loss on an $8.4 billion writedown. Citigroup wrote off $5.9 billion, then another $8 billion plus ⦠Goldmanās ānet revenueā from trading in mortgage-backed securities and other complex instruments was $31.2 billion for all of 2007. Last year, it contracted to $9.1 billion. Most stunningly, that trading turned negative in the fourth quarter. Its revenue was minus $4.5 billion. And thatās the top line for business activity. No surprise at all that Goldman reported its first-ever loss on the bottom line, at $2.1 billion for the quarter.
"This 'ZDNet Undercover' Webcast brings together a panel of experts to talk about and take questions on how Goldman Sachs marries human, technical and political capital to foresee and then move faster than rivals, when economic conditions change ... The webcast experts are:
Charles D. Ellis, Author, "The Partnership: The Making of Goldman Sachs,"
Robert Arvanitis, Principal, Risk Finance Advisors
Ron Papanek, Head of RiskMetrics Labs, RiskMetrics Group
And the event is chaired by Tom Steinert-Threlkeld, Blogger and Moderator for ZDNet Undercover Webcasts."
Goldmanās success dates back to the mid-1940ās, and is based on their recruiting policy which is far tougher and more rigorous than any other firm has been over time. Other firms may claim they hire top quartile or decile, but Goldman go for the top 5% and then sort them out over 2-3 years once on board. Thatās why they have had so many star players, such as Lloyd Blankfein, David Viniar, E. Gerald Corrigan and Hank Paulson.
They are also very disciplined. People at that level are going to normally have an ego, but no egoās are allowed at Goldman.Ć The word āIā is not allowed, itās always āweā. Theyāre not even allowed to see the eye doctor.
Itās all about teamwork too. They have all kinds of divisions, with two or three co-heads running each area simultaneously. Itās hard to make that work but, when it does, it works beautifully.
Thereās no hierarchy, and everyone is expected to take leadership in decisions that matter the most. The intensity with which they communicate all the time, 24*7, is amazing too. I mean everyone spends half an hour before sleeping and immediately upon waking doing voice mails and emails as their priority.Ć That's dedication.
How do you take this culture and communication capability, and then apply data systems to underpin it?
Well, flight simulators donāt make you a pilot and models donāt make you a financier. Thatās why Goldman tempers the financial models with business judgement, and the feedback models with cultural overlay. Those are two different models and both have strong human input to them.
Thatās why, according to the latest figures from Bloomberg, Citi has taken $84.4 billion in write-downs, Bank of America $40.2 billion, JPMorgan $29.5 Billion, Morgan Stanley $20.5 billion and $7.1 at Goldman so far.
Goldman Net Revenue from Trading:
2007:$31.2 billion
2008:$9.1 billion
4Q 2008: -$4.5 billion.
They also have a very strong capital commitments committee that is actually the most important decision making committee in Goldman.Ć You may think itās the Board, but itās this committee as they are responsible for keeping Goldman's capital afloat.
By way of example, on the day the markets dived 23% in a single day, every part of that negativity had been scoped out in a report that day and sent to the committee showing what could go right and what might work. Thatās a very different mindset to thinking about what could go wrong.
If you go to most trading, the reporting of most risk management is through the head of trading or the head of that business unit.Ć That creates a very inefficient system as a lower status individual ends up arguing with a higher status one, and you can guess how that works out.
Risk management needs to report directly to the top ā the CEO or CFO ā so that risk communication can happen without getting overridden by alpha male characters. Thatās the criticality of why Goldmanās team culture works.
From a technology perspective, Goldmanās key risk technology is based upon SecDB, a database and pricing system originally rolled out in 1996 which is now used across the firm in all their security trading businesses. They then use GridServer 4.0 to balance the load for risk calculations, as well as a whole range of relational databases developed in the 1990ās, to keep track of every product they sell to whom.
In the context of risk, they use a variety of methods to assess risks including historical events, predictable events, monetary values and various economic parameters. These are used for stress testing purposes against each instrument and trade.
The way you stress test is to shock an asset class with an event that has occurred in the past or that you might see happening in the future. An example might be a critical change to the dollar:yen. You throw that volatility against an asset class portfolio to see how that shock might trickle down through the portfolio, and the direct and indirect impacts.
Equally, for monetary values, you might shock a particular stock and see how that affects all of their stocks and bonds, and any derivatives and related instruments against that stock.
Finally, economic parameters would includes things such as a rally for the euro, gold pricing dipping and such like.
The thing is that everyone can get the models, but when you have the fastest car you need to bet on the driver, not the car and, in the case of the Lehman collapse, no driver had thought about the risks of who owned the collateral wrapped up in multiple Credit Default Swaps (CDS).
How do you get down and take collateral credit?
These were fields were not carried over between different instruments, which is why it was all buried below the line.Ć It also showed that no matter how good your model, sometimes you cannot build in an unanticipated risk impact. For example, you put in details of the previous CDO that a trade related to, but that information would not be carried over to the next trade.
Risk management also has two main components and often people look at one and not both. The two elements are the economics and the quants related to each instrument and trade.
Thereās a fundamental economic analysis to see whatās happening in a portfolio, and then a quantitative approach to risk management that should be independent of those fundamentals. You cannot do effective risk management when focusing on only one of those.
Thatās the reason why some of these models have failed. If you only look at one of these on their own in your risk framework, and just stress test one side of the equation with key drivers such as a house market failure, then you fail.
How Goldmanās work is that they:
1. Determine whatās in the pool of risks.
2. Rate the pieces.
3. Slice into ātranchesā of securities.
4. Figure out the expected default rate.
They then stress test these portfolios and investment and their ripple effects based upon industry, market, credit and business risks, and see the impact upon default rates, ROI, recovery rates and more.
So, for example, they would continually put in stress factors such as: what would happen if there were a major rise in default rates? What would our strategy be?
They would know the sensitivity of a whole portfolio for rises in credit defaults and the actions to take, along with the likelihood of price movement in the portfolio.
Thatās why itās critical to stress the portfolio based upon high default rates, even if you cannot see it happening, in order to know the ripple effects and price movements on other parts of the portfolio.
Some stress test by dropping prices or raising Value-at-Risk VAR) and doubling volatility, but you need to do all these together because price movements and volatility are independent of each other.Ć You need to know the ripple effects each have on the portfolio and that's what Goldman's do.
But if these systems are all the same, why would Goldmanās build their own rather than using a package?
Itās not that systems donāt matter. Of course you want the fastest car to win the race, but everyone will be pushing towards the upper quartile of performance. Therefore, what distinguishes you in the race is how you drive that car, not just having the fastest car.
Everyone will want the best systems, but it has to be the best systems that fit your culture. Thatās why you have to overlay human judgement, and allow traders to say that if the market is doing X and be doing Y, then give them the latitude to deal that way.Ć That's what Goldman builds into their systems, based upon a team culture.
Itās not only using the best models but knowing what the competition will do and being able to make judgement calls on that.
How will the other players react and can you anticipate that and get to the door first?
Most risk managers focus upon the details in the model but the top risk managers, such as Goldman's, focus upon the assumptions in the model. What are you assuming will hold true when you look at the outcomes of a decision and relying on that model?
The model will only work its outputs based upon those assumptions and this is why that is key, especially as the model works on correlations of risk that were input at a particular moment in time. Those correlations can change over time and you need to be aware of that, and change them accordingly.
There are a lot of components under assumptions that are direct drivers into the vulnerability of your risk structures.
Thatās why Goldman is winning out, because if you look at the Big Five investment banks ā Bear, Lehman, Merrill, Morgan Stanley and Goldman - and how this has played out, Goldman is the stand out (comparative results for third quarter 2007):
Goldman Sachs: $1 billion profit.
Merrill Lynch:$8.4 billion writedown.
Citigroup: $5.9 billion writedown.
Itās not to say they didnāt take a hit so if you cannot see the rabbit, look for the footprints in the snow.
Now we know Goldman are good but they may have taken some excess risks albeit by accident. For example, commercial banks talk about total accumulation to counterparties. Investment banks just talk about a trade and, as long as both sides offset, then youāre done.
Goldman over accumulated risks with AIG through CDS as a result. This footprint is shown by the fact that Goldman were far less leveraged than their competitors but still pricing competitively, so they must have had some off-balance sheet cover through CDS.
This would make up for the implicit versus explicit imbalance between prices.
Now, AIG was good for the counterparty risks because that meant Goldman did not have to put up collateral.
What probably then happened, when Lehmans collapsed, is that another part of AIG had collateral calls which triggered the domino effect.
Goldman Sachs, being in both the collaterals side and the CDS side, could very rapidly use their culture of fast communication across the internal human network to rapidly understand how those domino's were falling. They could see the dominoās start to fall and how, in a few days, it would hit elsewhere.
That gave them the critical jump to get out of the door fast before it hit their house.
Chris Skinner
www.thefinanser.co.uk