Emidio Sciulli joins Sandy Rattray on The CIO Agenda to discuss how investors can extract more alpha from their trading strategies.
With the rise of systematic trading and machine learning-driven investment strategies, investors are moving towards faster, higher turnover strategies. For those strategies, the measurement of these costs is crucial and potentially can cause some strategies to become unprofitable if not properly considered.
The easiest way to boost the alpha of a trading strategy is to reduce costs. But what is the best way to do so with multiple orders in the same direction? Emidio Sciulli joins Sandy Rattray on The CIO Agenda to discuss how investors can extract more alpha from their trading strategies.
Recording date: 29 June 2021
Note: This transcription was generated using a combination of speech recognition software and human transcribers and may contain errors. As a part of this process, this transcript has also been edited for clarity.
Sandy Rattray (00:05):
Hello, I'm Sandy Rattray, CIO at Man Group. Welcome to The CIO Agenda podcast. Today, we'll be talking about alpha opportunities and hidden costs in trading. And my guest is Emidio Sciulli, Head of Fast Trading Strategies, at Man AHL. Welcome Emidio.
Emidio Sciulli (00:22):
Thank you, Sandy.
Sandy Rattray (00:25):
So let's kick off just by exploring the alphas that exist in the short term. So what opportunities do exist for alpha and short-term strategies?
Emidio Sciulli (00:36):
It's a very interesting space the short-term space, and I think it's fair to say that there are plenty of opportunities, but they have changed over time. Historically short-term traders were much more similar to sped up versions of CTAs. While if we look at sort of the landscape now the top layers really look a lot more like a slowed down version of a high frequency period.
Sandy Rattray (01:03):
And so, and so let's just explain that a little bit more. So a high-frequency trader that would be a market maker effectively, or somebody operating like a market-maker and a CTA would have much slower signals. So you're saying that you think there's been a convergence or a movement for short-term strategies from being CTA like to being more market maker like.
Emidio Sciulli (01:26):
I think so in a way a CTAs they usually tend to be, first of all, from a data perspective, very much price driven the way they tend to sort of think about forecasting model, it's all about sort of forecasting trends in the market and the way they sort of trade those forecasts is to be very much liquidity takers. While if we sort of look at the market making space or the HFT space, the datasets are very different. So information on liquidity in the market, such as information on order books or the volume is a lot more important. If you look at the way they sort of forecast the market, it's a lot more about trying to capture the balance between buyers and sellers. And if you look at the way they trade, it's a lot more about mixing, taking and posting liquidities rather than just sort of taking liquidity. So if we look at sort of the short-term space, it used to look a lot more like a faster version of what CTAs used to do, but from a data modeling and journey perspective, why we feel look at them now, they look a lot more like a slowed down version of the market makers, where datasets and trading styles and signals they look a lot more like what is done by those players traditionally.
Sandy Rattray (02:32):
And if you look at the asset classes where these strategies are employed, I suppose, historically the HFTs really started in equities. Although now they're dominant in FX and in futures as well. Whilst the CTAs obviously started in futures and FX and some have moved into equities. So for you in very short-term strategies, are there some asset classes which are preferable to others?
Emidio Sciulli (03:03):
I think it's definitely a factor say that the more liquid the asset classes are the easier it is for short-term players to operate. And that's simply because of liquidity. The more liquidity the market is the easier it is to trade faster these series to not cause impact or having a strong footprint in the market. One thing that we've seen more recently is gradually short-term traders pushing the boundary on the liquidity space in asset classes. So asset classes that traditionally will be less liquid than not the traditional space of systematic payers. And definitely not the faster ones, such as, for example, credit or some of the cash fixed income instruments. They started becoming more interesting for short term traders. And the reason for that is because we have asset classes where the level of datasets and trading infrastructure available is sufficient enough to operate at the faster frequency and in a systematic way. But the overall amount of systematic players is not enough there to reduce the opportunity. So that makes it a very interesting environment for short-term traders.
Sandy Rattray (04:09):
Okay. And let's explore a little bit the types of models that you use. So certainly historically very fast traders really didn't want to talk about their models very much, and they were extremely secretive about them whilst slower traders tended to think they had more capacity. And so it would be a bit more upfront or open about what they were doing, what their models look like. So can you try and give us a flavor of what types of models end up in the very short term trading strategies?
Emidio Sciulli (04:43):
Sure. I think It's an interesting comparison. It's true, slow traders usually tend to be more open about the trading strategy and that's usually because capacity or market impact or risk in being sort of read by the counter practices is, is less of a, it's less of a worry. Faster traders there's definitely seemed to be more concerned about giving away. There are full at least the intuition that that office trying to capture, but at the very high level, I think there are for sure three types of models they're most short term traders will try to capture at least two types of intuitions that that those short-term traders will try and capitalise on. One is definitely reversion. So mean reversion in markets, it's a typical trading strategy that is played by the short term traders.
Emidio Sciulli (05:32):
The other one, it revolves around information around the order book. So anything that helps understanding imbalances in the market between buyers and sellers, it's a very, it's a very useful signal that's for, for short-term traders, because it allows them to capture on those market microstructure information that it's very important for them and very prevalent in their data sets. And then finally, the other one is also cross asset information. There is a lot of information between us the classes that can be used in order to forecast, for example, one of the class with another. And this is something that at the short-term frequency space is tend to be quite important and quite useful in forecasting what's ahead.
Sandy Rattray (06:13):
And if we dig a little bit into how you determine whether a short-term signal or model is likely to be of high quality or low-quality, how do you do that and how is that different to slow strategies?
Emidio Sciulli (06:26):
Yeah, that's not an easy, that's not an easy process. And I think in many ways it's similar to as lower signal to do research on slower signals. I think the main difference probably is on the, is on the focus. The reason on trading costs. I think as for those lower signals. It all starts with obviously having a solid intuition on why a signal work. So what is it capturing? What is it trying to get him paid for? And then obviously the actual research process themselves. It has to be very, very, robust, for example, how do we not minimise overfitting for example, by having a very robust sort of sampling sample or a sample process or, or, or length of data that is used to test hypothesis, robustness and stability tests are particularly important. And then finally, obviously a robust set of analytics that allows monitoring both on the research stage and then on the trading stage the difference between expectation and natural realised restart. So from this perspective is very similar to to doing research on slow trading strategy. What is very much different is the, is the importance of training cost. So looking at trading metrics and, and, and costs is, is a lot more important when doing research on faster trading strategies.
Sandy Rattray (07:38):
And we'll come to that because we need to spend, I think, a good part of our session talking about cost, but before we do that, people often believe that very short-term strategies have much higher Sharpe ratios than slower strategies. Is that your experience?
Emidio Sciulli (07:56):
I think it's fair to say that in general, a short term strategy should be more able to capture movements in the market compared to like a slower strategy, which means that you should have a Sharpe ratio. It generally tends to be the case. There are some differences between asset classes. So for example, you mentioned at the beginning that a lot of the short-term players at the beginning were on equities, and then it sort of moved into kind of FX and futures. And I think it's fair to say, for example, that in, in asset classes like equities, when there are more assets, a lot more idiosyncratic information that is specific to various assets, it's easier to generate for example, higher Sharpe ratio than compared to effects in futures where the assets are fewer and they tend to be driven by the same common macro economic factors.
Sandy Rattray (08:43):
Okay. That's a very helpful now let's just dig a little bit more into opportunities within markets and also where the, where the new opportunities lie. You touched earlier on, on credit. For example, clearly there are a lot of people who are players in short-term strategies, in futures, in FX and equities. So is it these newer markets, sometimes less liquid markets that really represent the big opportunity for you right now?
Emidio Sciulli (09:15):
I think less liquid market or in general markets where the, the presence of systematic players is less compared to, to to more sophisticated markets, such as for example, FX or futures or equities are one of the most interesting areas for, for short term players. And the reason for that is simply that ultimately capturing returns to the market. So by having an edge, an edge, which can be everything is set on data set, or a better ability to process data sets or better trading capability. So if you look at asset classes that are less liquid, less dominated by systematic players, such as for example, credit or some of the markets in the fixed income space or in alternative energies, there is a lot of data there. There is enough sort of technology and trading capabilities to allow systematic players to operate, but at the same time, the vast majority of traders are still not systematic and definitely not fast then makes it a very fertile ground for short term traders.
Sandy Rattray (10:16):
And maybe just to explore two other aspects of what might make markets interesting for short term traders. So firstly, do you think it's correct to say that short-term strategies are really long volatility they tend to do well when volatility levels are high that tend to do not so well when volatility levels are low, is that a fair characterisation of the, of the opportunity set? And if so, why, why is it there?
Emidio Sciulli (10:42):
To a certain extent. I think it's not always like that. And definitely not for all asset classes or for the type of signals, but in general, many asset classes that are traded by short-term traders and many signals, actually the best short-term traders they tend to capture on the tend to, they tend to capture large movements in markets that have been usually suddenly. So when there are periods of market volatility they usually just find a little more opportunities. And especially if the volatility is very is uncorrelated between markets that makes it even easier to sort of capture returns.
Sandy Rattray (11:16):
And how about the composition of, of players and market? You touched on this already, but, but the composition of the market in terms of concentrated a small number of dominant players, or a very large number of people active in market institutional versus retail mix, any features that tend to make the market more interesting for you?
Emidio Sciulli (11:42):
I think I would probably say two aspects are definitely important. The first aspect is one that we already sort of touched on at the high level, is this the, the split between discretionary or manual traders versus systematic or quant players. That definitely drives a lot of the opportunity set that simply because humans are better able to sort of process information, for example, where the data is patchy, hard to process or difficult to find. So those markets usually are sort of, you know, more reachable opportunities for a mono trader. For the faster player you want markets where there's a lot of information. It's a, it's very easy to access. And at the same time, the training infrastructure is very sophisticated. So that's definitely one aspect to step an extent, especially in equity markets the importance of retail versus institutional. It's also a big factor in in understanding profitabilities. So that definitely would have to be another one.
Sandy Rattray (12:43):
And let's, we're going to come to costs in a minute, but before we do that, one other thing I want to talk about was risk and how you think about risk and short-term strategies. So most risk models, and certainly the ones that you can buy off the shelf from third-party providers run with daily positions, they take the data in every day, and then they calculate some risk numbers out of that. That's clearly not appropriate for very short-term strategies because your positions are going to move around much more quickly than a daily model would allow for. So how do you think about risk and calculating risk for short term strategies?
Emidio Sciulli (13:22):
Yeah, that's a very good question, I think for risk systems, and I think more generally for many of the, of the components of an infrastructure for a short-term traders, it is really important to have an infrastructure that is intra day. It's also, it's a, it's quite specific for, for faster strategy. So risk systems definitely they need to be built allowing to sort of measure values, risk metrics intraday, also for the sort of trading infrastructure and ability to sort of innovate and adapt trades. Generally what you will find is that the, the top on the, definitely on the top end of the, of the faster trading space infrastructure and risk metrics and, and, and, and general sort of the whole kind of framework that is used to, to go from alpha to market, tends to be sort of built on purpose for, for, for those classes, strategies, and tend to be done in house rather than outsourced.
Sandy Rattray (14:14):
Right. Okay. Well, let, let's now move, you talked about costs and I think by which I think we're talking about trading costs firstly, just before we actually even get into really defining what you mean by that, I think many of our listeners would think that very short-term strategies wouldn't be paying costs, it'd be earning costs, they'd be earning the bid offer spread. They wouldn't be, wouldn't be paying impact costs. So but you are clearly not thinking that way. So, so can you help clarify that point? Why you think about costs when many people might think you that could be earning the spread?
Emidio Sciulli (14:52):
Yeah, that's a very, that's a very good question. I think it's, it's it's very much driven by whether the strategy is, is taking or if you're providing liquidity. If it's providing liquidity like the typical bread and butter market maker strategy that is trying to under spread. It's fair to say that costs are different than what you will expect from elite strategy. They sticking ingredient, which is basically crossing the spread in order to get into position. Strategies that provide liquidity, they tend to have less costs, so they still have an impact in the market just by virtue of, of interacting with other counterparties. But it's a, it's a different type of order money to them and, and type of cost and what you have in a strategy that is taking liquidity and, and, and impacting the market with this trading. Normally, when we worry about trading costs for faster strategy, we're more worried about strategies that are taking liquidity rather than strategies that are providing liquidity.
Sandy Rattray (15:42):
Okay. So that clarifies that. And then let's talk then about how you think about costs. Again, I suspect you think about cost very differently to a relatively slower investor or somebody taking positions with holding periods of weeks or months or even years. So with these shorter term strategies, how, how do you define costs and how do you think about them?
Emidio Sciulli (16:09):
I think the, the most important thing when thinking about costs, especially for faster strategies, but the same, to be honest, applied to certain extent, also to medium and slow frequency strategies is, is making sure that both visible and hidden costs are taken into account.
Sandy Rattray (16:27):
And so what do you mean by those? What, what is visible and hidden?
Emidio Sciulli (16:32):
The difference between both simply comes from the, from the correlation that we have between our trades. If I'm a portfolio manager and I have a view on a certain market let's say, I want to buy Euro dollar. I issued an a hundred million dollar trade, for example, and obviously any sensible execution, I'll go, we'll go and split that trading into smaller trades. Now the smaller trades that are part of the larger world, there are, are by structure structurally inter-correlated. And usually also the, the larger trade that the portfolio manager sent also tend to be auto, to correlate to the further trade the same portfolio manager will do. So when people normally talk about visible costs, they just refer to the difference between the price at which the portfolio manager decided to trade, so-called the season price and the actual average price of the values orders in the market. When we look at it then costs, we think about that in terms of the permanent shift that we caused in the market, maybe two of our correlated trading behaviour in the market.
Sandy Rattray (17:30):
Okay. And just to be absolutely clear by auto correlated, do you mean that if there was a trade on, in one direction in the previous period, then it's lately that the trade done in the current period is in the same direction, in the same instrument?
Emidio Sciulli (17:45):
That is right. And that comes from the fact that, you know, in this case, the portfolio manager, if he's buying one market, there is a highly chance that the way his building his forecast is, is gradually building up an exposure. So that's why the trade will be auto corrected, which means that if he buys now, it will probably buy late and so on, depending on his alpha.
Sandy Rattray (18:03):
Right. So your point is that when you have trades, which are auto correlated, where they're likely to be for several, it could be days, it could be 10 minute periods. It doesn't really matter, but many periods in succession in the same direction that people sometimes miss the price impact they cause because they tend to measure the cost from the beginning of each period, as opposed to from the beginning of the whole period. Is that correct?
Emidio Sciulli (18:29):
That's right. Yep.
Sandy Rattray (18:31):
Okay. So so then, so you have this hidden cost, which conventional cost models would miss convention cost models just say, well, you did you executed in this period? What was the price at the beginning of the period? What was the mid price at beginning of the period? What, what costs, what, what, what price did you execute at? And the difference between those two would be the cost on you're saying, if you have multiple trades over periods sequential periods, then you will miss this, this large effect from having orders in the same direction or auto correlated orders. So how do you calculate a true trading cost then when you have auto correlated orders?
Emidio Sciulli (19:13):
It's not simple because obviously it's measuring an impact that is by construction hidden, but there are a number of models available out there. And each of them with different pros and cons, I think one model that is particularly intuitive because it captured in a very simple way the essence of the problem, is the so-called Expected Future Flow Shortfall. And I think what will that model does in a, in a nutshell is to try to sum up the impact of the difference in price between two consecutive orders and the expected number of orders that we're going to have in the same direction. And obviously the crucial ask that to sort of [inaudible] in this model are how many orders in the future that I'm going to be aggregating on. And that's usually a function of the, of the alpha or eyes-on of the trading strategy that we were talking about.
Sandy Rattray (20:04):
Okay. And then if you split an order into smaller pieces, how does that impact all of this?
Emidio Sciulli (20:11):
Well, it's smaller orders in general help, in the sense that on the positive side of having small orders there, it's helpful because it makes orders less visible. That's obviously it's helpful from a, from an impact perspective, it's also makes it easier for your counterparty, for the market makers to internalise the order on their book. And at the same time also it makes it easy for them to manage the risk that their onboarding wants to trade. On the negative side I think that the problem with smaller orders that is when you have smaller orders that are auto correlated, because when you have a lot of small orders that are auto correlated that obviously that can cause impact as we just described. And also they can become very detectable. So HFTs or market makers, they can easily forecast what your next trade is going to be because they know that there is that correlation in your trading profile.
Sandy Rattray (21:00):
So what do you do about that?
Emidio Sciulli (21:03):
I think it's a balance. So you want to try to slice your trades in the, in the smallest possible chunks in order to minimize your impact, but you also want to make sure that your auto correlation profile of this trading behavior produces an overall impact in the market that is of the right size, given the alpha you're trying to capture.
Sandy Rattray (21:23):
Okay, well, let's move on and talk about models and just in little bit more detail. So many of our listeners may be familiar with the implementation shortfall approach or model which was developed by Andre Perold about 30 odd years ago now. So it's pretty well understood where essentially you look at your fill price when you're doing a trade. So what was your average price? And you compare it to the price before you started. It gets a bit more complicated in terms of how you define 'before you started', but we might just take the mid price before we started trading before we had the impact. It could also be of course, the moment when you decided you wanted to do a trade, but you and a number of colleagues have developed a model called the expected future flow shortfall. So what is the expected future flow shortfall model?
Emidio Sciulli (22:19):
So the expected future flow shortfall model is essentially a model that tries to augment implementation shortfall in order to account for hidden costs. So the idea is that the, the true short fall is made of the difference between the price between two consecutive orders and the, and the sum of all the expected orders in the same direction that we're going to put in the market.
Sandy Rattray (22:44):
Okay. And what, when do you use it? What types of situations is this most useful for?
Emidio Sciulli (22:51):
Well, in one way, you want to use it for every trading strategy, hidden costs are going to be paid by every trading strategy, whether it is fast or slow, but in practice, it makes a much bigger difference for trading strategy that are large with respect to the trading floor that they put into the market compared to market volumes. And, and it's the gift for strategy that they have a very auto-correlated trading profile, so faster trading strategy. They naturally tend to have an impact. That's why making sure that you measure it properly. It's crucial.
Sandy Rattray (23:22):
Okay. And they have a higher impact just because they're trading more often. Not, not because there are a large proportion of the volume. Typically,
Emidio Sciulli (23:31):
I think what matters a lot is the fact that they trade the trades faster and in a very auto correlated fashion.
Sandy Rattray (23:38):
Right. Okay. And then how does a model like the one you talked about respond to different market conditions, more volatile, less volatile, more liquidity in the market, less liquidity in the market. How does it adjust to those environments?
Emidio Sciulli (23:56):
I think that the main valuable there that drives impact is is market volume. So it's fair to say that the thing that is liquidity in the market the more, the more pronounced the impact will be well when a market is very deep there are lots of volume where they play the force, that the more, the less the impact will be in that market from trading.
Sandy Rattray (24:18):
Okay. Let's move on and talk about time horizons for orders. So the classic implementation short-fall model really sort of assumed that you knew what your, your execution timeframe was going to be. It's particularly well suited for. You're going to do an order across the whole day, and you have some price at the beginning of the day. It could be the open, it could be the price that you took your decision, and you compare your fill price to the starting price, whatever it was, but it assumed that you would be done within a day, and that you'd be happy being done within a day. So how should you think about a world where actually you're not sure how long your order is going to take, and you might have multiple different horizons depending on what the conditions look like.
Emidio Sciulli (25:06):
Yeah. So in terms of the, from the perspective of how do we model impact for, for that, it's a, it's about trying to understand what is the average holding period for the, for the signals. So we will know in advance how many orders in the same direction we will do, but but what we can do is to estimate the number, what is the average length of our, of our trading in the same direction of what is our average holding period. And that's what you usually use to kind of tune the parameter of the model to measure impact.
Sandy Rattray (25:39):
And so now let's move on and talk about how short-term alpha signals are combined with costs and in these shorter term trading strategies one level you might imagine that you simply say, well, you know, I want a high alpha strategy and it'll probably have high costs, but so long as the alpha is bigger than the cost, then it's a good strategy. If the alpha is lower than the cost, then it's a bad strategy. Is, is that how you think about it? Or is it more subtle than that?
Emidio Sciulli (26:10):
I think in a way it is as that, but in another way, it's a little bit more complex. I think what's more complex is that one wants to try to to optimise the alpha, the one can generate various horizons and, and the costs that it takes in order to build up the forecast that is that their alpha is trying to capture. So for example, you might have a signal that is very fast and obviously as just started that very expensive and it's time horizon is very, very short, but at the same time, you may have something a little bit slower, and honestly, they tend to be they tend to be fast. They tend to be less expert. Tend to be less, less expensive. And so the solution is not to put everything obviously into the very fast, higher, up signals to try to understand, for example, if the signal that is a little bit slower is giving me a clean forecast on a, on a slightly longer bigger, can I then use, for example, my fastest signals to support a modulating my trajectory in order to capture the alpha from broad signal.
Emidio Sciulli (27:07):
So I think that that interaction between signals or different horizon is actually quite the quite important in, in tuning the speed of the strategies.
Sandy Rattray (27:15):
Okay. So you need to model a lot of different horizons, maybe just again, for our audience. What, what are some examples of shorter term type signals versus relatively longer term, nothing as long-term in this area, but relative long-term some just examples of signals that you've experienced at a very short term versus a little bit longer.
Emidio Sciulli (27:40):
Yeah. So for example, the very fast extreme of buttons and one can capture you definitely have a very fast, for example, a market maker structure, basically. So there is lots of information in the market, micro, structural, the other book that, you know, to be captured that needs to be very fast. So this is or the booking balances, strict pressures trading balances. All of these is, is very, tends to be very, very fast. If you kind of, we've began to go on those lower end of the spectrum, relationships such as, for example, information between asset classes, they're sort of spillover from one to another, or sort of like faster version of trends tend to be much slower.
Sandy Rattray (28:21):
Okay. And so for each of these categories of the faster and the slower, then can you really measure the alpha decay clearly with any of these, you stretch your execution over a longer period, but if you're getting a lot of alpha decay over a longer period, the signal won't work anymore. So how do you measure the alpha decay with, with respect to time in other words, after you've generated the initial signal?
Emidio Sciulli (28:47):
That's not an easy task to do. I think that I have to ask this will be how you do it when you are simulating the strategy and how you do it when you're actually trading strategy and monitoring, whether they're outside the case, changing over time. I think that simulation stage, or when you're doing the research, it's slightly easier in the sense that what you can do is just to sort of check various performance metrics to assemble Sharpe ratio would be an obvious one and see how it changes. If you start lagging your signals, for example. So seeing that, for example, a signal that tends to have a pretty persistent Sharpe ratio and that is robust to lagging. That means that the alpha horizon is lower than, for example, a signal that in comparison tends to sort of, the key is sharp ratio very quickly, as soon as we start lagging it, once we start trading, obviously it's a little harder to settle, understand if alpha is decaying or not, for example.
Sandy Rattray (29:39):
And so for you as an investor, when you're trying to work out the, the right trajectory, do you do that by some sort of optimisation, or what do you do? How do you work out the right trading trajectory for different speed signals?
Emidio Sciulli (29:55):
So there are different approaches, which is that, that we, that we sort of use in market participants use. So one is much more based on sort of optimization. So you can obviously try to map out your signals to a number of forecast horizon, and then you can try to map out your trading costs or, or your cost of entering position to trades or by the sort of length. And obviously you can kind of optimise it, which has got those two metrics and any other metrics that you want to add such as the, is so correlation into like a, an objective function that you are essentially maximising. So this is definitely kind of one approach. And obviously the advantage of this is that it's very neat from a modeling perspective. This is obviously estimating all these quantities is a lot harder. The other kind of side of the approach is to have something a little bit more realistic, where you try to sort of map out your signals to various horizon and, and, and you try to map out styles which you want to trade by this horizon.
Emidio Sciulli (30:48):
And you sort of treat your signals in a more independent manner. The advantage here is that it's a lot easier to model those quantities a lot easier to estimate but you sacrifice a little bit the model the optimality of the model.
Sandy Rattray (31:02):
Okay. And then finally, when you're doing all this work, thinking about costs for quite fast strategies or very fast strategies, how does that help when thinking about more traditional slower strategies, which also of course have to execute in the market, does it help think about how to implement those strategies better?
Emidio Sciulli (31:22):
Yeah, it definitely does. And there are particular two aspects that are, that are very that are very sort of synergetic of, of the two. So the first one is obviously the infrastructure. So a lot of the, of the trading infrastructure, trading capabilities that one needs to build for faster trading strategies can be leveraged out completely by as lower trading strategy. And the advantage that you have, there is a substantial reduction in cost and impact of these of those of those medium, more slower frequency strategy where costs are less important than for faster strategy. We're still like a substantial part of, of the P and L. And the other aspect is of these, the ability to sort of combine faster signals with slower signals, so being able to sort of have a suite of fussing of the compliment, the trading of lower signals, and for example, modulating the floor, for example, accelerating entry to position when it's advantageous in the short term versus lowering down when it's less advantageous is the other areas where synergies are.
Sandy Rattray (32:17):
And maybe I could just explore that infrastructure word with you just a little bit more. Clearly, it's hard to make a slower infrastructure work fast. So something which could trade once a day is not going to do a very good job of trading intraday and something which trades maybe a few times a day is not going to do very well trading much faster or low latencies. So how do you think about the optimal infrastructure for fast strategies? How do you think about developing that? What does it look like? Can it run on coding languages that more traditional strategies use, or do you need to, do you need to run on a different infrastructure?
Emidio Sciulli (33:03):
Yes, that's true. There are lots of, there are lots of sort of aspects here. So I think from a perspective of sort of what are the crucial components, I think it is really important if you want to trade fastest strategy in a, in a, in a sort of, you know, top tier quality way is to have a very solid tree generation stage and a very solid execution stage origination stage is very much where all the sort of decision about how to sort of package signals into trades, how to sort of net the merge signals choosing which execution styles to attribute to every signals are made. And then the execution stages will actually then implement those decisions in the market. So by actually going and placing an order on a certain order the book, for example, or with a certain kind of practice on both components needs to be, needs to be there for a, for a tier one fast trading infrastructure from a perspective of startup platform, when you were mentioning trading language is obviously the execution world.
Emidio Sciulli (33:54):
It's, it's a lot more low-latency. So having languages there, programming languages and infrastructure that can cope with that level of latency is important. So usually, you know, Java or, or similar set of languages that are able to operate. The second speeds are the one that are more, more used while I think languages that are more useful research such as Python, for example, they can, they're more common in that, in the trade generation stage, where there is a lot more research and analysis. So they need to be done in a kind of dynamic way.
Sandy Rattray (34:25):
Okay. Wonderful. Well, so we've discussed today, really the, how high frequency trading and short-term trading strategies have converged to a significant degree that there isn't a separation in the same way as you might've seen historically between very fast strategies and slower strategies. It's more a continuous spectrum. And of course, we've talked about costs and how to think about costs in a faster world. One where you may often be doing the same order multiple times in the same direction and where your costs are likely to be quite a large portion of your alpha. So thank you for joining us today, Emidio and thank you to our audience for listening. You can follow The CIO Agenda on Apple Podcasts, Spotify, and other podcast platforms to receive each new episode. Thank you and goodbye.
Emidio Sciulli (35:17):
Thank you, Sandy.