The practice, as published
Everything in this section is a paraphrase in this site’s own words. Nothing is quoted. The works it comes from are listed in the next section.
- The work is divided on purpose.
- A portfolio manager decides what to hold. A dealing desk carries the decision out. A risk function, which reports to someone other than the manager, measures the exposure and enforces the limits. Compliance checks the rules; operations settle the trades. The point is that nobody checks their own work.
- Liquidity is asked about first.
- Before a large trade the desk asks how much of the thing changes hands in a day, what share of that the order would be, how deep the order book is, and how many days it would take to get out again. A position that takes weeks to sell is treated as a different risk from one that takes minutes.
- The cost of a decision is measured against the price when it was made.
- André Perold’s 1988 paper named the implementation shortfall: the gap between a paper portfolio that deals instantly at the decision price and the real one. It includes the spread, the movement of the price caused by the order itself, the drift while waiting, and the trades that were never completed.
- An order moves the price it deals at.
- Albert Kyle’s 1985 paper gave a model in which the price moves in proportion to the flow of orders, because other participants cannot tell whether the buyer knows something. A large order therefore pays more than the quoted price, and part of the movement stays after it is done.
- Fast is costly; slow is risky.
- Robert Almgren and Neil Chriss set out the trade-off in their paper of 2000. Dealing at once pays the most for moving the price. Dealing in small pieces over time pays less for that, and leaves the unfinished part exposed to whatever the market does meanwhile. A schedule is a chosen point between the two.
- Limits are written before the trade.
- Typical limits cap the size of one position, the exposure to one issuer, sector, country, currency or counterparty, the amount of borrowing, and the share of a portfolio that could not be sold quickly. Some are hard blocks in the dealing system. A breach is reported upwards; it is not left to the judgement of the person who caused it.
- Everything is reviewed afterwards.
- Dealing costs are compared with a benchmark after the event, results are broken down into their sources, and losses are examined for what the process missed. Regulators in many places require firms to have a policy on how they obtain the best result for clients when dealing, and to show that they follow it.
The sources
Named by title, author, publisher and year so that they can be found and checked. No links are given.
- The Implementation Shortfall: Paper versus RealityAndré F. Perold. The Journal of Portfolio Management, 1988.Used here for: The measure of dealing cost against the price at the moment of decision.
- Continuous Auctions and Insider TradingAlbert S. Kyle. Econometrica, 1985.Used here for: The model in which price moves in proportion to order flow.
- Optimal Execution of Portfolio TransactionsRobert Almgren and Neil Chriss. Journal of Risk, 2000.Used here for: The trade-off between the cost of dealing quickly and the risk of dealing slowly.
- Findings Regarding the Market Events of May 6, 2010Staffs of the U.S. Commodity Futures Trading Commission and the U.S. Securities and Exchange Commission. Report to the Joint Advisory Committee on Emerging Regulatory Issues, 30 September 2010.Used here for: The official account of one large sell order and how it was executed.
- What Happened to the Quants in August 2007?Amir E. Khandani and Andrew W. Lo. Working paper, 2007; published in the Journal of Investment Management.Used here for: An analysis of several days in which many similar portfolios lost money at once.
- Minimum capital requirements for market riskBasel Committee on Banking Supervision. Bank for International Settlements, January 2016.Used here for: The replacement of value at risk by expected shortfall in the rules for banks’ trading books.
A documented example: one order on 6 May 2010
The official report on the events of 6 May 2010 describes a single large order. It does not name the firm; it calls it a large fundamental trader, a mutual fund complex. On an afternoon when markets were already under strain, that trader began to sell 75,000 futures contracts on an American share index, worth about 4.1 billion dollars, to protect an existing holding of shares.
The order was given to an automated program with one instruction, as the report describes it: to sell at a rate equal to 9% of the volume traded in the previous minute, with no regard to price or to time. The report notes that the same trader had sold a similar amount earlier that year using a mix of manual dealing and programs that took price, time and volume into account, and that the earlier sale had taken more than five hours.
This one took about twenty minutes. As prices fell, fast-trading firms bought contracts and quickly sold them on to one another, which raised the traded volume without adding anyone willing to hold. The program read the higher volume as room to sell faster. The report’s point is that volume is not the same thing as liquidity.
The report’s weighting of this order has been disputed since, and in 2015 the American authorities brought charges of market manipulation against an individual trader in connection with the same day. What is not disputed is the description of the instruction, and it is a clear case of a schedule that followed one measure and ignored the price.
Where the process itself failed: August 2007 and the limits of a number
In the week of 6 August 2007 a number of funds that chose shares by statistical rules, holding some and selling others short, lost large amounts over a few days, out of proportion to anything in the wider market that week. The paper by Khandani and Lo, written shortly afterwards, suggests that many such funds held similar positions, that one or more of them was reducing its book quickly, and that the selling pushed prices against all the others at once.
Each fund had risk limits. The limits were set from each fund’s own history, which contained no record of what would happen if its neighbours all sold together. The authors are careful to say that their account is an inference from a simulated strategy and from public data, not from the funds’ own books.
The second failure is of a measure. For years the standard figure for market risk was value at risk: the loss that should not be exceeded on all but a small share of days. It says nothing about how bad the remaining days can be. After the crisis of 2008 the Basel Committee replaced it with expected shortfall in its rules for banks’ trading books, published in January 2016, for that reason.
The general lesson institutions draw is modest. A limit is only as good as what happens when it is reached, and a risk figure built from the past describes the past.
What can be studied with a small paper portfolio
A private account cannot have separate people for each job. What can be borrowed is the habit of writing the limits first and reviewing against them afterwards. The exercise uses a hypothetical account.
- Write three limits for a hypothetical account before anything else: the most that may be lost on one position, the most in positions that tend to move together, and the fall at which everything stops for review.
- In the Risk Room, replay one run of trades at five sizes on the sizing ladder, then use the correlation machine to see how several positions can behave as one.
- In the position size tool, work out the size that keeps one invented trade inside your first limit. Note how much smaller it is than the size the margin alone would allow.
- Record hypothetical trades in the journal for a month. At the end, play the part of the reviewer: count the entries that broke a limit, whatever their result. A breach that made money is still a breach.
- The Risk RoomRisk of ruin, the same trades at five sizes, and correlation, on invented figures.
- Position size toolThe size that fits a chosen loss, from your own inputs.
- The journalA record to review against the limits. It stays in your browser.
A limit that one person sets, watches and may waive is weaker than one enforced by somebody else. Writing it down helps. It does not make it independent.
What does not scale down to a private account
- Liquidity
- The institution’s problem is being too large for the market. A private order is usually small enough to be filled at the quoted price, so slicing it serves no purpose. The private trader’s costs are different ones: a wider spread in proportion, and financing.
- Derivatives and venue access
- An institution deals directly on exchanges and other venues, through several brokers, with programs built for the purpose and commission rates it has negotiated. A private account deals through one provider at that provider’s price.
- Financing terms
- Institutions borrow and lend securities through prime brokers and the repurchase market on negotiated terms. A private account’s financing is whatever its provider charges.
- Information and staff
- The process depends on separate people: one decides, another deals, a third can say no. A private trader is all three, and nobody can overrule a decision made at the wrong moment.
- Operational resources
- Measuring dealing costs, running stress tests and monitoring limits through the day take data, systems and staff that cost more than most private accounts hold.
- Tax treatment and time horizon
- A pension fund or an insurer invests against obligations decades away and is often taxed differently from a private person. Both facts shape what it can sensibly hold.
- A leveraged CFD is dealt differently
- A contract for difference is an agreement with a provider at the provider’s quoted price. There is no order book to work an order through, and nothing is owned. The exposure is usually leveraged, usually carries a financing charge and can be closed out by a margin call.
Questions people ask
- What is implementation shortfall?
- It is the difference between what a portfolio would have earned if every decision had been carried out instantly at the price when it was made, and what it actually earned. It gathers the spread, the price movement caused by the order, the drift while the order was being worked and the cost of trades that were not completed. The term comes from André Perold’s 1988 paper.
- Why do institutions split large orders into smaller ones?
- Because a large order sent at once uses up the quantity available at the best prices and has to deal at worse ones. Splitting it gives the market time to refill between pieces, which lowers that cost. The price of doing so is time: the unfinished part is exposed to whatever the market does meanwhile, and others may notice the pattern.
- What does an independent risk team do?
- It measures the exposure of the portfolios, checks them against limits agreed in advance, runs tests of what would happen in a severe market, and reports to senior management separately from the people taking the decisions. Its independence is the point: it does not depend on the judgement or the pay of those whose positions it measures.
The words on this page
A summary for study of practice described in published papers and official reports, in this site’s own words. It is not endorsed by, or connected with, any institution described. Nobody at GIO4X has access to their portfolios. Studying a method does not reproduce its results, and a leveraged CFD is not ownership of an investment and does not behave like one. Educational information, not investment advice or a recommendation to trade.