Finextra quotes Tabb as saying there are 55+ venues competing for (US) equities e-trading business … which sounds like the ideal landscape in which to be selling smart order routing ..the tools to pick out the right venues for the order in question. I wonder whether there is a market for buyside smart order routing in fixed income space too; and whether this technology has potential to change the general dealer reluctance to permission a client on more than one platform for a particular product class?
I have commented in previous posts that the next generation of pricing systems will be one of the foundations for innovation in the etrading space, so heres a quick checkpoint; if your bond pricing system is still largely restricted to instrument on Y axis and its attributes in tabular form on X, then check out the demos at lab49 to start worrying about how soon you’re going to have to start again from scratch. 🙂
EFN revealing (?) that Managers are suffering as quant similarities are revealed….A quantitative manager said: “The falls are across the quant management board and that shows that, although the equity models may be superficially different, really they are variations on the same theme. Everyone has been taking long positions on value stocks and short positions on growth stocks, and now they are all making losses.”
so are the really smart cookies finding alternate venues and products in which to prove and capitalise on their investment/trading/positioning skills?
Following on from previous post about algos and FIX in Europe, Finextra shows FPL take another step forward in standardising the protocol for algorithmic order types… So far 15 industry participants, including six of the world’s largest broker-dealers, have particpiated in testing of the new language. Each firm has published samples of their algorithmic trading strategies in the new XML format, ensuring that all their current and near-term algorithmic trading strategies can be expressed in the new format.
EFN says fund managers in Europe have been slow to embrace algorithmic trading (12% of trading volume at asset managers, doubling since last year) compared with their US counterparts; blaming poor technology, and saying that while MiFID may eventually encourage algorithmic trading, fund managers’ time and resources are focused on other projects … one recurring theme with higher priority being to implement FIX.
With the industry still being at the nuts-and-bolts connectivity stage and hence without experience to fully trust electronic trading, limited uptake of algorithms, and another EFN article saying that dark pools of liquidity are relatively untapped by European Fund Managers, I can’t help thinking we still have an enormously long way to go to reach a post-MiFID vision of buyside connecting to multiple liquidity pools electronically and using algorithms to compare all in search of best execution.
But what’s the alternative? An ALLQ type view of the different pricing available for a particular stock line, and a myriad of venue-specific manual order entry forms as part of their desktop? With FIX implementation being top priority that’s clearly not the desired route. But for those of us with a fixed income background, does that kind of setup ring any bells?
I was thinking about the use of news feeds in algorithms again, and wondered why Reuters and Dow Jones and all the other vendors trying to flog these feeds aren’t out there letting people use the tagging / weighting / scoring just to filter what appears in different panes in their market data terminals. Letting everybody via the market data gui see how they can filter the crap and only see important stuff, or filter all the bad news in order to see a rosy view of the world, or whatever, if only just to build faith in the provider’s story scoring. Let people begin to play with the scoring system (such as the ‘weakness’ comment in ukalgotrader’s News feeds into algorithms? Lets make algos work first!) via the market data gui and see where it goes?
A topic that came up at the Financialnews Sellside Technology seminar was the ever-increasing demand for proximity technology – where systems are being placed as close to the exchange as possible in order to avoid latency. In the context of “what’s coming in the future”; one question posed (by Dave Cliff the “ziptrader”) was how long before exchanges charge to host a firm’s algorithms coded on silicon? So rather than connecting a machine with the shortest possible cable between itself and the exchange, a card containing the algos is instead plugged into the exchange machine itself? An interesting future prospect of revenue stream for an exchange?
A question raised was whether that materially impacts the ability for providers to deploy algos to customers immediately; and on that I thought “immediately” was interesting terminology.. (same thought applies to ullink framework that avoids the need for ullink software release just to get new algos out of the door). Are customers willing and able to utilise new algo’s immediately? Surely there is is a time lag before a new algo is used materially at most customer-side. Time for the customer to understand what the algo is trying to achieve and how, then for this to be cleared by the “fat controller” and worked into the [new] trading strategy.
How long is the time lag? Depends how much trust the customer has in the algo provider (eg sellside) actually getting this sort of stuff right each time. Also trust that the software provider (OMS/EMS/intermediary who exposes the route into the algo to customer) isnt going to muck it up along the way, and trust that it is possible to get a comprehensive audit of what where and when to show that an algo acted in accordance with it’s claimed design – whether for best execution certification or when there are occasions that the results are vastly different than expected.
Much of this comes down to trust that the testing has been done and the tests covered everything they should. Surely this isnt materially different if the algos are on silicon plugged into the exchange rather than software somewhere in the pipe between customer and exchange? And given the resource that could be given to this, if there is a need to establish a lifecycle which includes pressing algos onto a card which can then be plugged into the exchange machine, is this really going to slow algo deployment down to the extent that this impacts sellside ability to grab competetive advantage?
Again on the future of OMS/ECNs… I saw a press release a while ago re ULLINK’s UL REACH thing that apparently lets sellside publish their algorithms to buy-side community so they are made available “instantly” (meaning without having to wait for a new ullink software release).
Why don’t the FI ECNs build their GUIs to support transactions being built from generic building blocks of functions and fields? Build the equivalent of their existing etrading offering as the generic functionality using those building blocks, and then shift ‘development’ of anything over and above that to the sellside who wish to support a particular type of transaction.
As long as dealers wish to support it (and so develop it), this removes the constraint that the ECNs place on what can be transacted on the platform. With the right building blocks could dealers support things such as customers electronically unwinding existing swaps once non-standard, or etrading butterflies and curves and so on ? .. the things some big buyside say they require before they will do any of their swaps business electronically. In terms of value-add the ECN becomes “the library” of core etrading functions as well as the connectivity between customer+liquidity provider – not a bad position to be in to ensure they remain in the future landscape.
As the OMS guys are already coding dealer-specific functions (algos) into their gui, if the ECNs continue to hard-code their gui-per and so constrain the product offering then where does that leave them in the future?
The machines are taking over the world!
Bloomberg article on “machine outwitting humans” and AI in the algo space says Language represents one of the biggest gulfs between human and computer intelligence, Dhar says. Closing that divide would mean big money for Wall Street, he says. Unlike computers, human traders and money managers can glimpse a CEO on television or glance at news reports and sense whether news is good or bad for a stock. In conversation, a person’s vocal tone or inflection can alter — or even reverse — the meaning of words. Let’s say you ask a trader if he thinks U.S. stocks are cheap and he responds, “Yeah, right.” Does he mean stocks are inexpensive or, sarcastically, just the opposite? What matters is not just what people say, but how they say it. Traders also have a feel for what other investors are thinking, so they can make educated guesses about how people will react.
So there you have it – the human value-add comes down to making “educated guesses”.
What if the other end of the conversation (/information flow) is a machine, and there is no vocal tone to interpret? Are we expecting machines to also utilise sarcasm in their responses?
Link to the FT mandate panel discussion about the uptake of algorithmic trading, and the lack of education holding back implementation of algos.
- General agreement that at present algos are tactical – to build and trade out of a position over a day or so, rather than algos to make long term asset allocation and run stock positions. – Is the next wave to be more about long term decision making?
- Cliff Warner (London & Capital) says the job of the buy-side desk is to maintain the beta, to mitigate the risk and therefore to get me the best price on the market, whereas the alpha is generated from his fund managers. – thus algorithms are perceived a risk to buyside execution desk
- Mention of using FIX for algo’s. – still work to do
I see algo trading is in the uk telegraph today
Dow Jones has launched a data feed that delivers news items in XML format which can be rapidly processed by electronic trading programs(so same sort of offering that Reuters recently launched). It says its new Elementized News Feed delivers data directly to quantitative-analysis models and automated trading programs and is designed for both buy and sell side firms involved in algorithmic trading, quantitative analysis and execution management. The ultra-low-latency service delivers economic and corporate news in precise and discrete elements in XML-tagged fields. This eliminates the need to parse unstructured text, allowing trading models and computer programs to process, interpret and take action on breaking news in milliseconds.
If you’re thinking about using this sort of thing, isnt the first question how much trust you are going to put in their tagging? whats the error rate of items – how many are edited and retagged (after your algo machine has done something on the basis of the first tag)? Who does the tagging? – if it’s the journalist do you know for sure all of their journalists will tag in consistent fashion? if its a separate “tagging” desk doing this then how much latency is added? And of course is the backtesting history tagged by hand by the same people and with the same time pressures so the history is consistent?
Electronic sizes increasing. Greater percentage of business going electronic. But with more algos being used to price each inquiry .. and trying to keep the price as wide as circumstances and order history shows the customer will probably accept at that time, unless the bank has a very clear policy that the machine’s price is the only price available at that point for that customer, then with the knowledge that there is occasionally still some margin (even if that margin is not profit per-se), surely the future is for buyside to continue phone in their important orders so they can beg, steal, borrow, threaten, ask for a favour just this once for a price improvement?
Some axe grinding in response to Waratahpost geeing up his readers to challenge the status quo-
– Banks make prices and Funds take prices
– Algo trading isn’t for Fixed Income
There is no fundamental barrier between making prices and taking prices, but those making a living market making are by nature going to need to be good at sticking their prices out in whatever shop front and enticing customers to trade, whereas those on the customer side need to have honed their skills to know where to shop for the best opportunities to get the deal they want. My post a while ago suggested “executable bid lists” might give the ability for top tier customers to send their inventory requests in executable form to dealers – I expect we will enjoy an increase in this activity as direct customer to dealer connectivity increases, and this is offered by dealers as a value-add for their top customers.
The difficulties in a market maker doing this for all of their customers in processing unsolicited requests from many thousand customers at one time is purely “scale”. You can’t just present each request as a popup msg to the trader to manually decide (like many did in the early days of Tradeweb’s mybid/myoffer in the $ treasury market). The machine needs to be able to work out whether to accept or not – which gets us onto Algos.
The other point about this is why is the customer sending in their price. I can see a join between an OMS generating the orders needed to execute a particular move in a portfolio (adjusting overall duration or whatever), but would you want customers to have a gui on one of the platforms to just send you their own price … without your machine on the receiving end being able to instantly auto-reject all the crap?
So Algos in the fixed income space. You will find algos in fixed income, in pricing engines. The easiest place to start with this is responding to RFQ (because at that point you have a reasonably narrow set of data to crunch). So your algos look at all of the relevant data related to the inquiry in question and return the “right” price. Of course once you trust your algo to be able to do that, letting it loose market making on your order driven markets is just scaling this up – after all, it’s just a broader selection of inventory, still with a specific type of counterparty at the other end. If you need to handle executable bid-lists/IoIs where customers send in what they want and the price they’ll do it you’d need the RFQ processing and the ability to accept when the customers price is “close enough”, determined by how much you want/dont want the position.
Conventional wisdom suggests that the demand for customer-facing algos in the fixed income space will increase after they sort out their algos for FX. I also see the MiFID triggered changes in the exchange markets putting OTC closer on the algo radar too, as the re-engineering to make the existing (exchange focused) algos able to deal with fragmented liquidity does mean they are architecturally similar to what would be required in OTC markets. Why would a client not wish to utilise algorithms to systematically go-get their best execution? Of course the fundamental problem then will be whether the fixed income quotes the algos are relying on are truly executable or not – which is another item I’ve already added to waratah’s list.