Anyone knows what are the hottest research areas in algorithmic trading?Scott

momentum based strategies on very liquid indices are floating about the street at the moment (marketing blurb warning - usual crap ... "aim to extract 'strategic' arb from indices" - "transparent hedge-fund lite approach" - "have low correlation to other asset classes enhancing efficient frontier") - trading signal usually a exponentially weighted oneAlso seen some interesting carry type algorithmic trading (bucket of X yield curves ... short Y [lot less that X] flattest, long Y steppest)there are other types i've seen - again all relatively simple (almost niave). While backtesting on these strategies looks good ... doesn't it always I have my doubts about these strat

Do you mean by research, research in academia or research in companies or both?

I saw some dispersion trades attempted in algorithmic framework.. not sure how it works though

Trading algorithim often means different things on the sell-side versus the buy-side. Buy-side trading algos are "strategic" (deciding what to buy/sell and when) while sell-side trading algos are "tactical" (trying to get the best price for a buy/sell order). Tactical trading algos are a big research area because these systems are automating the jobs of sell-side traders who used to handle block trades (1 block = 10,000 shares), while providing clients with low costs, fast execution of large trades, and good buy/sell prices.The main research areas in sell-side trading algo are pre-trade and post-trade analytics (predicting and measuring the market impact of a large trade), which is powered by statistical analysis of tick data. This is the area that gets most attention from academics. The central concept here is to minumize permanent and temporary market impact by slicing an order over time. A very profitable research area for the sell-side is VWAP (Volume Weighted Average Price) trading. It works like this. Let us imagine that you are a brokerage client who wants to sell 100,000 shares, one morning. Your broker agrees to buy the 100,000 shares from you at the Volume Weighted Average Price. As soon as the market opens, the broker's VWAP server starts trying to match the fill price with the current VWAP. At end of day you will find out what the VWAP was for that day, and the broker has guranteed to pay you the day's VWAP for your 100,000 shares. Now, the broker makes money by using a sophisticated trading algorithim which can sell the shares at a price that is better than VWAP. So the broker's profit is the difference between the day's VWAP and the actual fill price (plus trading commisions, of course). Portfolio managers like to buy and sell stocks at VWAP mainly because VWAP is a standard benchmark in the industry. If you do a big trade your boss will generally measure your execution performance by comparing your fill price with the day's VWAP. If you don't want to risk getting in trouble with your boss for doing an ill-timed trade then you send a VWAP order to your broker. Yet another research area in sell-side algos is smart order routing and algorithmic sweeping of dark liquidity pools. Many people do not realize that the market structure in the United States is very fragmented. There are not just two trading venues (NASDAQ, NYSE), in actual fact there are at least 100 trading venues. Meaning, if you want to sell IBM stock, you can go to 100 different electronic trading networks which are collectively known as Alternative Trading Systems. Among the Alternative Trading Systems is the so-called dark liquidity pool (also known as a crossing network). It is called a dark pool because the order book is hidden from the participants (i.e. you don't know what the bid/ask in the pool is); this allows traders to park huge blocks of shares in the order book without worrying about sending out a signal to market participants. The dark pool matches limit orders and executes them at the mid-point of the bid/ask price quoted in the normal exchange. There are many different dark liquidity pools in the US, and in order to "source liquidity" (get your order filled) you have to go around "pinging" these dark pools, until you hit a match. There are algorithims which perform this sweeping in an intelligent manner by estimating the probability of hitting a match in a given dark pool. This dark pool sweeping algo is integrated into smart order routing systems. The other function of smart order routing is to get the best price in the least amount of time while incurring the least amount of exchange-related costs. Reg NMS (SEC regulation concerning the National Market System) means that brokers are legally required to execute client orders at the best price (a practice known as best execution), which has caused a big push towards smart routing technologies. The most intriguiging research area is "stealth trading algos" and "opportunistic trading algos," which is about concealing signals that might be seen if trading is done in a less subtle way, and trying to time trades during the day so that they are filled at the best price. Stealth trading algos go by names such as "guerrilla algo" and "ambush algo" and they are powered by historical tick data and real-time analysis of the order book (guerrilla algos, for example, use supply/demand imbalances in the order book as triggers). I'm not sure how much academic research has gone into this, but these kinds of products are now offered by all the major brokers. ITG (Investment Technology Group) was a big pioneer in this area, and they've also been running a dark liquidity pool called POSIT for over 10 years now.

Very good summary, though focused on equities. FX and Fixed Income trading are copying some ideas from the equities side, and vice-versa. Some banks are now offering execution algorithms in fx. The algorthms aren't the same since the information you get varies from market to market, but conceptually most of the algorithms are pretty much the same. Different liquidity pools are a given in fx, and some of the ideas on how to get big orders done in fx are flowing back into equities.

Question: Where (if anywhere) are the papers on algorithmic trading being published? I get the sense that this is a "if I tell you, I'd have to kill you" field right now.

Journal of Trading http://www.iijournals.com/JOT/

Robert Almgren hass published some papers.

Are you aware of interest rate instruments liquid enough and traded electronically to make algo trading possible ?

What is the difference between the following (or do they all mean the same thing)?- Algorithmic trading- High frequency trading- Statistical arbitrage- Systematic trading

I'll take a stab at this. I would say that, generally, 1) and 4) are synonomous. High Frequency trading refers to intradaytrading, on the time scale of minutes and even seconds. Statistical arbitrage is a broad term forbuy-side methodologies that attempt to generate alpha by looking for statistical patterns in prices,and then taking bets on their recurrence.

ThanksCan we say for instance that high freq trading belongs to algorithmic trading?Can we further say that both are part of stat arb?If not, do the techniques used differ much or is there a great deal of overlap?As an example, is signal processing and filtering useful mostly for high freq trading more than alg trading?

These 4 terms you refer to do not constitute a hierarchical classification so you should not expect more relations between those than there is.Really there is no more to understand than what Skyhawk has just replied to you.To answer specifically to your latest questions:1)HF trading can be described as algo trading if it is implemented using an algorithm (which is most often the case).2)HF and algo trading will be described as stat arb only if part of the trading uses some statistical signals (which will often but not always be the case).Also these terms might primarily refer to different things to different people. For instance, a lot of people use "stat arb" to refer to longer term bets (compared to HF trading).So do not assume too much on the meaning of these words...

Isn't part of the problem that the folks in marketing tend to take over any popular term and slather it on everything? I wonder what fraction of "statistical arbitrage" funds and algorithms obey the original definition of arbitrage in being risk-free and market neutral?I suspect the term has come to mean any trading strategy, directional or otherwise based on statistics...As for signal processing and filtering, that's useful and used on all timescales. Signal processing, at least what was borrowed from EE, gets a bit tricky at the highest frequencies because: 1) the methods generally assume either discrete equally-spaced time steps or continuous time with time being an error-free independent variable and 2) the algorithms don't have any notion of an interval or spread in the true value of the data other than to discard the spread as "noise." Throw in a real-time stream of bids and asks where the time-stamps may be out-of-order, irregularly spaced, or wrong and the data needs a bunch of preprocessing to make it pretty for signal processing. Signal processing is good stuff, but, like anything, comes with assumptions about the data, generating processes, and noise processes.