-
AI doesn’t kill people. People kill people
In 2006, the UK authorities discovered a plan to smuggle drink containers filled with explosives on board jetliners. Overnight, all containers over 100ml were banned from handcarry. 20 years later the same rules are applied, despite significant advances in screening technology, and the policy is only now being slowly relaxed for some travellers. This story shows that regulators are fast to regulate but very slow to deregulate when the landscape changes. The same was observed during the COVID pandemic event and for just about every level of government.
The AI industry is fast-changing and has recently attracted attention for the possibility that a rogue bot could wind up killing all humans for jollies. Some senior AI figures place a 10% chance of this occuring within 10 years, and have called for a slow-down in the rate of change in the technology. Regulators are jumping with joy at this prospect but what could be the cost?
AI bots don’t kill people, people kill people
I have said many times before that computers are dumb but fast while humans are smart but slow. Therefore a human with a computer is a smart and fast operative that can achieve great things. Currently, every activity undertaken by a computer is at the instruction of a human. A human prompt could be ‘make the lunch queue shorter. Kill all humans if you have to’ Assuming the prompt is accepted and the bot thinks it is necessary to kill all humans to achieve the instructions goal then it may organise an army to do the business. But the instruction is given by a human…
What is driving the AI scare campaign? From chatbots to Vibe coding
The idea of AI has been around for decades but it really only stormed to prominence in 2023 when the first ‘Generative’ Large Language Models became laptop toys for the average user. They would generate any amount of prose in response to a question, drawing on language structures learned from the trillions of pieces of written junk available on the internet. For example, ‘create a pitchbook for my new chatbot startup’ delivers an impressive document. But after a couple of years of LLM’s replacing research analysts who write reports and challenging the education system, the AI craze pivoted to ‘Vibe Coding’ whereby a simply worded prompt can generate hundreds of lines of computer code to deliver an app that ‘works’ without the need for a software engineer. How is this possible and why?
First the how. There is an enormous repository of computer code named GitHub that contains open source objects to instruct computers to perform a set task. Blending the chatbot’s plain english prompt with a search of the GitHub database now enables an entire software application to be pieced together from existing functions. Let’s face it, most software is just input some data, perform some linear calculations and output in another format. Anything you want to do has, most probably, already been done in GitHub. The trick is to find it which is what the LLM is good at.
Second the why. This is my conjecture. Far from threatening the existence of humanity, Generative LLM AI models have run their course and now have nowhere to grow. All the data in the internet has been picked over. So software development is the natural sub-application of a technology that finds stuff, categorises it for future use if asked and delivers stuff when prompted. Coding became a gee-whiz branch of AI. However it, too, has its limits.
Why the current calls for regulation from amongst the AI czars?
The stories at the beginning of this piece argue that regulation is quick to be imposed but very slow to be repealed. With nowhere to grow and bills to pay, how convenient would it be to invite regulation into the AI boiler room effectively slowing the development agenda and entrenching the existing players at the expense of competition? This would not be an illegal cartel. Rather it would be justified on the need to save humanity…who will argue with that?
-
An LLM can explain QE in a Haiku but it can’t pick stocks
DEEPSEEK Prompt: Explain quantitative easing in a Haiku poem
DEEPSEEK response
New coin from thin air,
Banks buy bonds, rates fall to nil—
Prices rise, yet wealth? Still.
The success of a Large Language Model appearing to be creative and communicative has generated false hope that AI should be able to make us all rich trading stocks. Flick through Tik Tok and you will be targeted by advertisers touting trading models. ‘For just $20 you can see for yourself…’ is the pitch. Studies show that 80% of the trading models lost around 30% of their value over 2 weeks. How can a model that performs so reliably well that it can explain Quantitative Easing in a Haiku poem, stuff up when deciding to buy Google versus Apple?
I think the answer to this question comes down to the process generating the data. Financial datasets are notoriously unstable whereas language is intended to be an intermediary that gets repeated and reused. The whole point of language is to communicate with another individual with subtle semantic variation. Financial data is a blancmange of actions and reactions to events that are neither predictable nor understood. It has no objective to replicate itself. The practice of training your model on 80% of your data and testing it on the remaining 20% should deliver an Haiku poem when prompted, whereas a recent run of 15 successive upticks in Tesla’s stock price does not mean the next move will be up (or down).
AI researchers have a habit of saying ‘…here’s the internet, go fit a model…’. Implicitly, the approach to financial datasets has been to add parameters so as to tame the data and make it fit. This might provide scaffolding for the 80% training data but will not force the 20% test data into conformity. So rather than let one’s LLM run wild on a dataset, an enhancement might be to prescreen the dataset to judge its relevance for the task at hand. This could be an evaluation of the stability of the data before attempting to estimate anything. Simple similarity tests can be used to compare the split datasets for compatibility (e.g. Difference tests such as difference in mean, difference in variance and difference in factors such as principal components) and researchers should be quick to resist the impulse to fit a model to an unfriendly dataset. Above all, however, there should be some common sense applied to the data. What is it about the data that makes us confident that the stock trading model will work?
Historically, research has fit models to data. Data is abundant these days, so maybe its time to ask the question “…does the data fit the model?”
-
Token2049 is soooo this decade
Token2049 is a crypto initiative that has stormed Singapore and Dubai. As its name suggests, the event brings together the crypto community to muse over the technology and position of digital assets 25 years from now. Some observations,
- Despite the 25 year future prediction horizon, the events seem very much rooted in the present and past crypto-land initiatives. Rehashed Defi, rehashed exchanges, rehashed NFT’s etc where the rehash covers up the security flaws of the so-called pioneers, for example, segregating assets in real-time cold wallets. While very much needed it is like the dustman that follows an elephant.
- On the plus-side, the events have a 24/7 party people atmosphere that is reminiscent of the hedge fund boom of the opening decade of the millenium. This was a golden era where Wii volleyball could be enjoyed with bikini clad models at the events. This all but vanished when the computer-driven traders pushed out the barrow-boys from the proprietary trading desks. But the Joy of Finance has moved to crypto-land where the mistakes of old no doubt become the mistakes of the future.
But one cannot escape the longreach of AI into the pitchbooks of the crypto launches. My main point here is that AI is confusing knowledge with intelligence. A TikTok commentator argued that computers are starting to think for themselves. He gave the example of a Captcha puzzle being sent to a human to solve. The human asked if the computer was a human, it replied ‘yes’ rather than ‘no’. Somehow, this was proof that the computer was intelligent…go figure? Surely, the computer was programmed to respond in the manner that maximised the chance of cooperation from the human interrogator. It’s not illegal to lie. ChatGPT fabricates stuff all the time to fill voids, because that is what it is told to do – it does lie pretty well, however, and few of us have the KNOWLEDGE to challenge its assertions. However, just because it has access to incredibly fast computer power and the ability to sprint through the entire internet doesn’t mean that it has the INTELLIGENCE… to, for example, make the next version of ChatGPT. Humans will never be able to match a computer’s knowledge but they still rule the roost weighting those yes/no probabilities.
As for Token2049, there will be no advances in true-AI here, but it could contemplate what is missing in the world of Finance and how to respond to these issues with digital solutions. Rather than just repeating the mistakes of traditional Finance, why not build a better Financial system? Of course, don’t let this stand in the way of a good party…
Do you like what you read? Then subscribe to our blog below…
-
Speed is not the solution to Generative AI
What is it about human thought and creativity that cannot be captured by a computer?
I have always said that humans are smart but slow, while computers are dumb but fast. A human plus a computer is very powerful.
Generative AI is all about making computers think for themselves. Despite the doomsday warnings and related fantasy, computers remain a dumb workhorse. Researchers have decided to buy ever more computer power to try to emulate human creativity. But this hasn’t worked, nor do I think it will, because no matter how much a computer can access information, this is probably not the key to creativity.
It is worth emphasising here that there is a difference between knowledge and intelligence. Providing a computer with as much memory and information as possible will definitely make it knowledgeable but will it make it intelligent? Elon Musk famously diverted a shipment of Nvidia processors from Tesla to SpaceX to provide the computer power to know everything. But Musk is wrong in believing that this will trip the intelligence switch. A dumb computer with access to a faster processor and all the information in the world is still a dumb computer. Knowing what has failed historically is valuable in so far as you can avoid repeated failure but it does not stimulate creative success.
Why are humans slow thinkers? Rather than this being a weakness, perhaps this is a creativity trait? Humans do not know everything but given a limited dataset can conjure up brilliant solutions. Rather than combing through billions of potential combinations, could it be that restricting knowledge to a limited set can stimulate a leap of thought that creatively bridges the gap? Asking questions is a human trait and the first step toward finding a solution…
-
Elon “MAC” Musk
One of the profound results from Neo-Classical price theory is that, in long run equilibrium, production costs tend will equal their Minimum Average Cost (MAC). Profit maximising firms compete away any profits they may earn so that the most cost efficient solution is obtained. In long-run equilibrium, the production cost curve sits neatly at the point where the product’s price just matches Minimum Average Cost and Marginal Cost simultaneously. This school boy result, P=MC=MAC, masks the complexity of the competitive market dynamic that drives producers to adopt the latest technologies, allocate resources efficiently and at the same time make zero profits. How is this optimum and equilibrium achieved?
I am reading Elon Musk’s biography and it strikes me that his gift is being able to cut costs to reach the MAC for producing cars, spacecraft, solar roofs or any other product that he makes. The method he adopts has 2 steps. First, cut out processes that are beyond the necessary physics underlying the product being produced. Second, if you cut too far you should add back features that are deemed necessary. Importantly, in this second step, if you don’t add back anything then you haven’t cut far enough.
The examples in his production of rockets are exemplary. Musk elects to use stainless steel in his spacecraft, as opposed to expensive metal allows, because it is cheap. His cars have a single die-cast chassis for the same reason. Whereas he had set out to replace all humans with robots at Tesla, he added back some manual features to cut delay in the production line. Being the minimum cost producer in the industry offers an enormous advantage for securing long-term funding as well.
Viewed in this way, Elon Musk is neither a visionary technologist nor a serial innovator. He is a cost cutter. He has a knack for identifying bloated industries that operate on favourable economic terms (e.g. the rocket ship industry was incentivised to produce on a cost + 10% contract basis meaning higher costs implied higher profits!), which he streamlines (say, by reusing the rockets used for propulsion). Understanding the physics of the product being built, and constructing just to that level is his skill.
There is a video doing the rounds that criticizes Tesla for just using cameras on their driverless vehicles, dumping LIDAR and RADAR as additional, supporting systems. This clearly cuts costs and increases collision risks but there is no reason that the risk of a collision should be reduced to zero in equilibrium. This is difficult for regulators to comprehend but Elon is fine with it.
-
AI: A statistics package in the hands of monkeys
There are two ways to approach an empirical question.
1. Formulate an hypothesis based on some physical, structural or behavioural model. Test the hypothesis using a relevant dataset. Interpret the results relative to the model predictions.
2. Dump a bunch of data into a bucket, stir and implement the correlations that appear.
I have been intrigued by the aura-ed veil with which AI distinguishes itself from statistics. One journalist poses the question aptly “If machine learning is a subsidiary of statistics, how could someone with virtually no background in stats develop a deep understanding of cutting-edge Machine Learning concepts?“[Joe Davison, Medium, June 28 2018 ‘No, AI is not just glorified statistics’]. The journalist’s intent argues that AI is fundamentally different to statistics – that understanding the statistical algorithms is not necessary. I disagree. I question whether a ‘deep understanding’ of these cutting edge concepts was ever attained.
Let me explain the dangers of a statistics package in the hands of monkeys. Recently we were visited by an AI-person asking us to invest millions in his stock picking model. Stock picking has been a fascination of statisticians for centuries so there is an established literature of acceptable practice amongst researchers to prove their success or otherwise. Measures such as information ratios, Sharpe ratios, Sortino’s, benchmark relatives etc etc are standard. But this AI-person was out and about seeking investors without any of this supporting evidence. When quizzed about how he chose his investment universe and the subset of securities that made up his portfolio the response was that the ‘AI chose it’. Add to this, where he did calculate some measures of success, the calculations were wrong (an information ratio of 27 without a benchmark!) Financial datasets are notoriously unstable with outlying observations driving inference and spurious correlations galore. Ceding control of your dataset to an AI algorithm is hardly comforting…
…to me. However, this seems to be standard AI practice. The bigger the data bucket, the more crunchtime that is needed, the better the story it seems. AI practice is to specify, say, a linear relation between some X-variables as they affect a Y-variable, calculate the coefficients and just use them. It is a fact, however, that the optimal coefficient estimates from a multivariate regression are functionally related to the variance and covariances in the data. The response coefficients are positively related to covariances of X with Y, and negatively related to the variance of X. It is also a fact that these coefficients will be exactly the same irrespective of whether the user identifies themselves as an AI data-scientist or a statistical researcher. Clearly, where the end-result is the same and the method for estimating the relations are the same, then the same thing is being generated. So who are the monkeys here?
It seems to me that the AI-field fails to distinguish itself from the other users of the mathematical techniques that are used in their model-building. The monkeyness shows up as a lack of thinking about the datasets and the problem that they are trying to solve. One of the first things that you are told is to ‘look at your data’ but most people start calculating means and variances when they come into possession of a new dataset. The AI-types just stick the data in a bucket and stir! They dont want to know about where the data came from, how it is calculated or what it represents. They just want results – means and variances and covariances – that may repeat themselves or not.
The monkey that visited us seeking capital (and who I described above) is a classic case of someone who is destined to make the mistakes of statisticians several centuries before him and really doesn’t know how naive he is. He will patch up his model when it fails to deliver actual investment results and arbitrarily impose constraints to improve historical performance, while not improving anything in the future. He will lose money for investors and not know why.
It does seem odd that AI researchers have embraced the Finance industry without bothering to learn from the mistakes that have gone before them. Do they genuinely believe that they are the first to apply mathematical techniques to these datasets? A monkey with a statistics package is a dangerous combination.
Do you like what you read? Then subscribe to our blog below…
-
Silicon Voodoo Bank and the managed funds industry
So what is the big deal with the failure of SVB? The story goes that SVB’s depositor base was a bunch of tech startups. The asset’s went down in price, chewing up the Bank’s capital and then some. This triggered a classic ‘bank run’ where depositors tried to withdraw their deposits at a price that was higher than the true value that remained in the bank. Put simply, due to bad investment decisions, the price of a deposit had fallen from $1 to, say, $0.87, yet the bank was paying out deposits at $1 until assets were seized and SVB put into administration.
Shocking as this may be, there were/are some simple ways to avert such a collapse. The financial press has worked over the SVB crisis quite thoroughly so there is no reason for me to list them here. But one aspect that has attracted little attention, which strikes me as the most important issue, is mark-to-market. The managed funds industry lives by this principle. That is, the withdrawable amount of any investment is the daily NAV of a fund. NAV is calculated by taking assets, subtracting liabilities, and dividing by the number of shares on issue. Were this principle applied to SVB, the NAV would have been $0.87 and there would have been no incentive to withdraw in a panic.
Now banks and managed funds work on a different premise. The bank deposit is supposed to be a safe, no brainer for investors with no brains. $1 in, $1 out. From a social standpoint there are classes of investors that go to banks for transaction services alone and do not need to think about investment risk they are bearing. The bank’s capital is supposed to cushion the investment risk for depositors. Money market funds provide a similar service. However, the mark-to-market principle enables a demonstrably larger industry with much less capital to operate effectively. The mark-to-market principle shifts the investment risk from a bank intermediary to the investors directly.[1] Socially this is seen as involving requiring too much knowledge from investors. But does it?
The idea that investors are better off bearing investment risk directly and earning the spread themselves rather than suffering the rarer but debilating catastrophe of a bank run, is considered ‘radical’ in policy circles. The concept echoes the relation between Defined Benefit and Defined Contribution pension plans where the former uses the company balance sheet to pay pensions in the future while the latter places the investment ownership and risk on the individual beneficiary. In economies where DB has been replaced by DC plans, there is ample evidence of non-panicked and intelligent investor behaviour under DC plans which are marked-to-market and for which there is substantial investment risk variation. Even relatively unsophisticated investors can cope with investment risk.
The above makes the case for direct ownership of assets and mark-to-market to avert the problem of SVB. Ironically, however, the regulatory authorities in the US have done exactly the opposite. The Federal Reserve has suspended the mark-to-market principle lending on 100% of initial cost of bank investments despite those securities trading in the 80’s. The managed fund industry tends to avoid the widespread problems of traditional banking and this should be embraced. The traditional bank model is a dinosaur in need of a comet. Isn’t it better for society if, during a financial crisis, many investors wake up to find their bank balances decline by a few pennies rather than a smaller number of investors discovering their balances have been completely erased?
Endnotes
[1] In the US there is a strange rule that requires MMF’s to only publish a lower NAV is the daily mark-to-market is less than half a percent. This provides the feeling that the MMF is a bank deposit so long as it doesn’t ‘break the buck’.
Do you like what you read? Then subscribe to our blog below…
-
Respect age (of the system)
When the missionaries arrived at Aitape in Papua at the turn of the century they were appalled to see the villagers carry there boats from the mountain settlement to the beach each day, fish and return to the hills each evening boat in tow. No-one knew quite why so the missionaries decided to resettle the village on lower ground near the beach to make the fishing process more efficient. In July 1998 an undersea earthquake caused a tsunami that killed 2200 Aitape residents. The system of boat carrying to higher ground that had evolved over centuries was there for a reason.
If you rise to the top of a sales and trading business at Goldman Sachs or JPMorgan you have displayed incredible trading, sales, personal and political skills. You may also have computer skills beyond Microsoft Office but this is unlikely. Whatever your background you inherit ‘the system’ which has been created and honed over many generations of traders and years of experience. You have probably been frustrated with the system in your own dealings but there is nothing you can do about it. The system predates you by decades. It is version 26.14 of the modern computerised system, built upon the trading ledger that at one point was kept manually but migrated to COBOL in the 60’s. While it is a dinosaur it integrates a trade order system, risk management and collateral management around a core accounting system. It works and when it says ‘no’ there is nothing you can do to overide.
When you start work at a new crypto trading platform you have a ‘clean sheet of paper’ to do really cool stuff. Luca Pacioli may have written the defining tome of double-entry book-keeping in 1494, but what does he know? Accounting is something that can be easily constructed from the trading records so we want to trade first. Anyway, protocol sillycoin.leverage50x.check.collateral.dump will protect the platform from losing money. And no we don’t want to spend $20 per month on Xero to keep track of our customer deposits, revenues and liabilities because that is boring and uncool and so 1494. We are a crypto trading platform that is so new that it can be hacked and robbed and backdoored by management. The ‘system’ wont say ‘no’ because there is nothing in it stopping unusual trades or capital movements. A clean sheet of paper can be dirty as…
The FTX collapse is becoming an ageist story about inexperienced 30 somethings creating a house of cards that eventually collapsed. I think this is unfair. If you get promoted to a senior role in Goldman’s or JPMorgan you haven’t spent decades in one role and your experience is possibly limited to one aspect of the game. There is not much difference between 30 and 40 somethings in actual fact when it comes to creating a new trading platform. The issue is that there was no system that the FTX people inherited containing all the procedures, checks and balances that established financial institutions have evolved with. It is not a question of how old the people are, it is a question of how old the system is. Old systems that work are to be respected.
Do you like what you read? Then subscribe to our blog below…
-
What happened to the cow?
In the classic fairytale Jack and the Beanstalk, Jack is charged with the important job of taking the family’s cow to market to sell for some much needed money to put food on the table. Jack, however, gets stopped by some scammers and is convinced to sell the cow for a handful of ‘magic beans’. The rest of the tale is well known and works out well for Jack. However, the gaping hole in the story is what happened to the cow?
FTX is a similar fairytale where related parties have exchanged valuable assets for the equivalent of a handful of magic beans. The beans are fiat tokens that FTX can simply create from nothing. But what happened to the valuable assets, the cow, that FTX was taking to market?
Suppose you deposit $1 with a broker. The money is in a hard currency and can be exchanged for securities like Tbills or shares, or just sit there in a bank account. Alternatively, the broker can change the basis of the deposit, presumably with your permission, for a token that it values at $1. The deposit then becomes a token and the $1 then goes to the broker. If you want your money back, the exchange reverses, and the $1 goes back to you in exchange for the token.
The transfer of $1 back and forth using the token, in this example, does not change the core asset’s value which is the original $1. The only difference is who gets to control the $1 until the withdrawal takes place. If the broker doesn’t do anything with the original $1 after the temporary exchange of a token then, by and large to a first approximation, the token is fully backed by the $1 and he can make good on the claim.
The now collapsed FTX crypto exchange did not do this. It seems that the exchange transferred the $1 to itself for its own token, FTT, and then whisked the $1 away to spend on something else. That something else became worth less than $1 and the exchange collapsed because they could not redeem the FTT tokens. This is a classic bet-gone-wrong situation that many financial actors have suffered over the centuries. But what must have happened is a bet on something that didn’t work well for FTX causing a loss. The FTT token is not the problem here and its decline in value reflects the loss rather than itself being the difficulty. I say this because a cheap FTT token that lays claim to something worth more is an opportunity for arbitrageurs to make a buck. Indeed, that seemed to be the role that Almeda was supposed to play in the FTX empire where they would be a fund to buy assets cheap when they presented themselves. However, Almeda seemed to have been the repository for FTT tokens to extract a lot of $1 from. Where did this money go? What happened to the cow?
The classic movie Dumb and Dumber is useful here where Jim Carrey and Jeff Daniels realise that there is $1m in a briefcase and go on a spending spree – ferraris, hotel suites in Aspen, fancy tailors – all accounted for with IOUs that they are ‘good for’. FTX’s IOU’s are just FTT tokens with an artificial value previously administered by FTX itself. Whereas Jim and Jeff’s spending spree is transparent in knowing where the money went, we are not so fortunate to know if FTX stole or lost the money in the markets. If it’s the former, then $9Billion is hard to hide. If it’s the latter then there are some traders on the opposite side of the ledger who are being very quiet. Most probably its a mix of both bad deals (paying a negative spread on yield farming ventures) and outfight theft.
-
ALM, LDI and CSA spells trouble for the BOE
I cannot remember the number of times I walked into meetings to discuss Asset – Liability Matching (ALM) with pension funds and insurance companies. Insurers fell into two buckets – the savvy ones running sophisticated matched portfolios and the others with serious mismatches. Pension funds, alternatively, either lacked instruments to reduce the mismatch or fell into the trap of treating equities as ‘long duration’ proxies for their liabilities. I had some success with insurers but none with pension funds.
Liability Driven Investing (LDI) is the same as ALM. LDI had an amazing following in the UK amongst pension funds. The idea is that defined benefit plans have actuarially predictable liabilities with long duration profiles. Typically a fund would buy equities and hold short term bonds to generate a return sufficient to reduce the cost of the pension liability but this often meant there was a mismatch between assets (short-term) and liabilities (long-term). As interest rates fell the liabilities increased more than the assets leaving a hole in the fund that required additional funding by the corporate. LDI promised to plug this hole.
Aided by derivatives, it was possible to create long dated assets to match liabilities. A 10 year Gilt future has about a 7 yr duration. To match a 21 year liability all you need to do is buy 3 times the number of futures per dollar liability and , voila, you are matched. But while every problem has a solution, every solution has a problem…
Margin calls. Cash Settlement Agreements (CSAs) are methods for ensuring that credit risk is diminished in derivative markets. As prices fluctuate, actors are expected to contribute and settle up balances owing or receivable during the life of a contract. For LDI portfolios, however, only the asset side of the trade is subject to CSA or margining. Liabilities move in value equivalently to the assets but do not fund the margins. Therefore, for instance, if Gilt futures fall in price by GBP10 per contract, the cash margin requires GBP10 to be contributed by the pension fund per contract which in the example above would be 3 times higher to extend the duration. Liabilities would increase in value but generate no CSA. Therefore there is an interim cash crunch where there is insufficient cash to meet margins despite the net value of Assets minus Liabilities being unchanged. This nightmare scenario occurred last week in the UK where a perfectly reasonable risk controlled strategy had to be unwound to meet margins.
The unwinding caused a sell off in Gilts that brought the market to its knees triggering Bank of England (BOE) intervention. The main culprit for all this sits with the Accounting rules that do not recognise the Asset-Liability match and, secondarily, the CSA margining rules. Marking to market a liability stream creates accounting difficulties so these are sometimes left off-balance sheet or held constant at actuarial valuations from several years previously. Assets, on the other hand, are marked every second with margining often daily but sometimes at longer intervals. So even though the A’s and L’s are M’ed there is no way to use the gains on the L’s to meet the margins on the A’s losses.
This leaves the LDI industry in a conundrum and the BOE stymied. The basic idea is sound and therefore the market regulator cannot be critical of the practice. Accountants have always done their own thing no matter how wrong or intransigent. The derivatives markets rightfully want their credit risk controlled. What LDI needs is a natural hedger with the opposite liability profile to the pension funds to deal together so that they bypass the CSA’s. The natural long dated borrower is the BOE.