5.2 Categorising algorithms
Based on fundamental mechanisms
- Schedule-driven: follow historical volume profile
- Opportunistic: react to favorable market conditions
- Evaluative: represent the middle ground between the above two extremes. At the macro level, behave in a more schedule-driven fashion, at the micro-level, focus on balancing the trade-off between cost and risk.
Focuses more on how algorithms are implemented, whereas a trader or investor makes their decisions based on set objectives or goals.
- Impact-driven
- Cost-driven
- Opportunistic
| Type | Key focus | Algorithms | Dynamic Benchmark | Pre-determine Benchmark | Price Sensitivity | Volume Sensitivity |
|---|---|---|---|---|---|---|
| Impact-driven | Time | TWAP | ✓ | |||
| Impact-driven | Volume | VWAP | ✓ | |||
| Impact-driven | Volume | POV | ✓ | • | ||
| Impact-driven | Impact | Minimal impact | ✓ | ◦ | ◦ | |
| Cost-driven | Price/Risk | Implementation Shortfall | ✓ | ◦ | ◦ | |
| Cost-driven | Price/Risk | Adaptive Shortfall | ✓ | • | ◦ | |
| Cost-driven | Price/Risk | Market On Close | ✓ | ◦ | ◦ | |
| Opportunistic | Price | Price Inline | ✓ | • | ◦ | |
| Opportunistic | Liquidity | Liquidity-driven | ✓ | ◦ | ◦ | |
| Opportunistic | Ratio/Spread | Pair/Spread trading | ✓ | • |
• often ◦ sometimes
Impact-driven algorithms aim to minimize the overall market impact. Cost-driven algorithms try to reduce the overall trading costs. Opportunistic algorithms take advantage whenener the market conditions are favourable.
5.3 Common features of algorithms
Algorithm parameters
Generic parameters represent common details, such as when to start and stop and whether to enforce a limit price, etc. The following list summarises most of the parameters that commonly supported, although various names may be used:
Start/End Times, Duration, Must-be-filled, Execution style, Limit price, Volume limit(maximum), Volume limit(minimum), Volume limit(child), Auctions.
5.4 Impact-driven algorithms
Impact-driven algorithms evolved from simple order slicing strategies. By splitting larger orders into smaller child orders, they try to reduce the effect trading has on the asset's price, and so minimize the overall market impact costs.
The average price-based algorithms, namely TWAP and VWAP, represent the first generation of impact-driven algorithms. Although intended to minimize impact costs, their main focus is their respective benchmarks. These are predominantly schedule-based algorithms, and so they usually track statically created trajectories with little or no sensitivity to conditions such as price or volume. They aim to completely execute the order within the given timeframe, irrespective of market conditions.
Time Weighted Average Price(TWAP)
Basic Mechanism
The simplest version of a TWAP algorithm is based on time slicing.
Common variations
Some versions may allow an additional factor to be applied to the trading schedule, which tilts it and so making it trade either more aggressively or more passively. A more aggressive stance will issue more orders early on, helping reduce timing risk, whilst a more passive approach should result in lower market impact costs.
A price adaptive TWAP algorithm might even adjust its schedule dynamically based on the market price(although arguably this makes it a simple price inline algorithm).
Special Parameters
Tracking, Interval frequency
Volume Weighted Average Price(VWAP)
Basic Mechanism
Modern VWAP algorithms often incorporate complex logic to determine whether they may get ahead of their schedule, and how best to catch up if they are behind their target. Overall, their performance is based on how well they track the target, but also on how well they predict market volume.
Common variations
The dependence on historical data means that VWAP algorithms can be vulnerable to sudden shifts in trading volume or liquidity. These can cause considerable deviations from VWAP. Therefore, some versions may also monitor current market conditions. Effectively, this makes them a hybrid between a static VWAP algorithm and the more dynamic volume participation approach.
Some Variants may offer a more adaptive approach that tracks short-term price and volume trends and dynamically adjusts their target execution profile accordingly. Though, such customizations mean they are not really true VWAP algorithms anymore.
Special Parameters
Tracking, Start time/End time, Trending/Tilting,
A more detailed example is given in Kissell and Glantz(2003).
Percent of Volume(POV)
Percent of volume(POV) algorithms "go along" with the market volume; hence, they are also sometimes called volume inline, participation, target volume or follow algorithms.
Basic Mechanism
Unlike algorithms, such as TWAP and VWAP, where a trading schedule may be predetermined, for POV algorithms the trading schedule is dynamically determined. The algorithm tries to participate in the market at a given rate, in proportion to the market volume.
Highlights a common concern about POV algorithms: If several are competing for an illiquid asset, they could start driving each other on, like two bidders in an auction desperate to buy the last lot. Potentially, this could lead to a significant market impact. To try to protect from such situations firm price limits may be applied to POV orders since the algorithm itself has no inbuilt price sensitivity.
Another consideration is how the POV algorithm actually responds to reported volume. If it simply splits a new child order each time there is a trade then the resultant trading pattern is nearly as predictable as simple order slicing, as Tom Middleton(2005) highlights. To prevent such signaling risk we can keep track of the target volume and trade more periodically. Trading in this fashion allows us to vary the order placement, and adjust the aggressiveness of our execution tactics based on how far behind we are.
Most POV algorithms use a variety of safeguards to prevent themselves from chasing volume spikes, such as comparing the target trade size with the currently available volume on the order book. Some versions may allow control over the market volume calculation, either to exclude block prints or set a maximum permissible trade size.
Common variations
Some POV algorithms may incorporate forecasting to try and better anticipate the upcoming trading volume. Generally, such approaches are based on a mixture of historical volume profile, current observed volume, and quantitative analytics.
Price adaptive versions of POV algorithms are also starting to appear. These adjust the participation rate based on how the current market price compares to a benchmark. Some variants can even adapt to the relative price changes for other assets, such as sector or market indices or even ETFs.
Corporate buyback programs in the U.S. often use the "safe harbor" provision granted by SEC's Rule 10h-18, which protects issuers against liability for market manipulation. This mandates strict timing, price, and volume conditions, designed to minimize the market impact of repurchases. Some brokers/vendors provide dedicated corporate buyback algorithms that adhere to these requirements. Often these are in fact based on modified POV algorithms. One of the key changes is to add a price condition that ensures the algorithm does not issue orders that end up making the market, so they do not allow their child orders to be priced more than the last trade or the current best bid.
Specail Parameters
Participation rate, Tracking, Volume filters, Start time/End time, Must-be-filled, Limit price, Execution style
Minimal impact
Minimal impact algorithms represent the next logical progression from VWAP and POV algorithms. Rather than seeking to track a market-driven benchmark, they focus solely on minimizing market impact.
Signaling risk is an important consideration for these algorithms. This represents the potential losses due to information that our trading pattern replays to the other market participants. It is dependent on both our order size and the asset's liquidity. Thus, algorithms focus on taking advantage of the facilities offered by "dark pool" ATSs and broker's internal crossing networks as well as using hidden order types to reduce this risk.
Basic Mechanism
The simplest version of a minimal impact algorithm is to route the entire order to a "dark pool" ATS and just leave it there. Although the actual hit ratios on some ATSs can be quite low, so many algorithms may also work a small portion of the order separately to ensure a reasonable level of execution.
Common variations
Some versions may incorporate models to estimate the probability of being filled on the ATS, using this to determine how much of the order should be left there.
Special Parameters
Visibility, Must-be-filled
5.5 Cost-driven algorithms
Cost-driven algorithms seek to reduce the overall transaction costs. These are much more than just commissions and spread. Implicit costs such as market impact and timing risk are important components of the overall cost.
To minimize the overall transaction costs, we need to somehow strike a balance between market impact and the overall exposure to timing risk. As what Kissell and Glantz(2003) refer to as the trader's dilemma: Trading too aggressively may result in considerable market impact, whilst trading too passively incurs timing risk. To strike the right balance, we must also take into account the investor's level of urgency or risk aversion.
Implementation shortfall represents a purely cost-driven algorithm. It seeks to minimize the shortfall between the average trade price and the assigned benchmark, which should reflect the investor's decision price.
Adaptive shortfall algorithms are just more opportunistic derivatives of implementation shortfall. In general, they are more price-sensitive, although liquidity-driven variants are also starting to appear.
Market-on-close algorithms strive to beat an undetermined benchmark, namely the future closing price. Unlike TWAP and VWAP where the average evolves through the day, it is harder to predict where the closing price will actually be. Again, the key goal for this algorithm is to strike a balance between impact cost and timing risk. Effectively, it is the reverse of implementation shortfall: Instead of determining an optimal end time, we need to calculate the optimal starting time.
Implementation Shortfall(IS)
Basic Mechanism
IS algorithms aim to achieve an average execution price that minimizes the shortfall when compared with the decision price(default the mid-price). As Middleton(2005) reports, some of these algorithms are essentially enhanced versions of VWAP or POV algorithms. These use cost models to determine an optimal trading horizon, which is incorporated as either a model determined end time or an optimal participation rate. Modern versions have often been built from the ground up, to react more opportunistically to price and liquidity.
Common variations
Adapting to market conditions, such as liquidity or price has become a key focus for many algorithms. Price adaptive implementation shortfall algorithms have become sufficiently important that the following sub-section is dedicated to them. IS algorithms which have access to "dark pools" of liquidity are also available.
Volatility is another key factor for IS algorithms. Some versions actually adjust to changes in short-term price volatility; When the volatility falls, they can afford to trade more passively. Other variants may also use alternative estimates for volatility. For example, Instinet's Wizard algorithm uses the implied volatility from the options markets as a more accurate gauge of intraday volatility.
Special Parameters
Benchmark price, Risk aversion, Execution style, End time, Limit price, Volume limits
Adaptive Shortfall(AS)
Adaptive shortfall represents a new subclass of algorithms derived from implementation shortfall. The adaptive moniker refers to the addition of adaptive behavior; predominantly this is a reaction to the market price. Effectively, this means price adaptive shortfall algorithms behave more opportunistically.
An AIM(aggressive in-the-money) strategy assumes that trends are short-lived and will soon revert, whereas a PIM(passive in-the-money) approach relies on the trend persists. Thus, AIM trading will immediately try to take advantage of any favorable prices and hope that unfavorable prices will revert. Conversely, PIM trading will react aggressively to stem any potential losses from an unfavorable trend, but it will be passive during favorable trends, hoping to achieve further price improvement.
Overall, Kissell, Freyre-Sanders, and Carrie(2005) found that the AIM strategy achieves better prices, but at a slightly higher risk. The PIM strategy is quicker to cut losses short so achieves a slightly lower risk, but may achieve poorer prices than AIM.
Common variations
Just as these algorithms react to the market price, we could also add behavior that reacts to changes in the order book depth or overall liquidity measures. Increasingly, they are also accessing liquidity from "dark pool" ATSs.
Special Parameters
Adaptation type(AIM, PIM)
Market Close(MC)
The close price is often used for marking to market, and calculating net asset values and daily returns/profit and loss. Hence, many institutions are still interested in the close price as a benchmark. Though, trading at the close can be costly. David Cushing and Ananth Madhavan(2001) found prices are more sensitive to order flows at the close.
Basic Mechanism
Many market close algorithms determine an optimal trading horizon using quantitative models, which incorporate estimates of an asset's volatility, and trading volume. So just as IS algorithms determine an optimal end time, market close algorithms can calculate an optimal start time, as Middleton(2005) points out.
Common variations
Some algorithms are now being termed departure price algorithms. These provide a more generic mechanism that will trade to any given end time, rather than just the market close.
Special Parameters
Risk aversion, End time, Auction participation
5.6 Opportunistic algorithms
Opportunistic algorithms have evolved from a range of trading strategies. They all take advantage of favorable market conditions, whether this is based on price, liquidity, or another factor.
Price inline(PI)
A price inline algorithm adapts its trading to the market price in a similar way t how POV algorithms adjust to market volume.
Basic Mechanism
A price inline algorithm consists of a basic trading mechanism combined with the price adaptive functionality. Hence, it could be based on a static VWAP algorithm or a more dynamic percent of volume approach.
Common variations
some versions may even allow the price adaptations to be based on the price of other assets, sectors, market indices, or even ETFs.
Special Parameters
Adaptation type, Benchmark price, Participation rate, Participation adjustment, Price levels.
Liquidity-driven
Highly liquid assets tend to trade in greater volume, and so often will have lower spreads and more depth available. Whereas for illiquid assets, activity is much lower, the spread higher and the order book may have little visible depth. So finding liquidity is an important means of reducing costs, particularly for illiquid assets.
Liquidity aggregation can be as simple as summing the available orders at each price point across all the different venues. For assets such as futures or options, this might also mean accounting for liquidity "implied" from more complex spread strategies. Venues have their own fee structures, so the algorithm needs to consider this when comparing prices. Latency is also an issue since any time lag between our orders being sent and processed could significantly reduce the probability of execution.
Basic Mechanism
Liquidity is closely related to market depth and price. Therefore, a liquidity-seeking algorithm will react strongest when there is plenty of market depth and the price is favorable.
Common variations
Liquidity-driven algorithms are often used in fragmented markets. Clients may want their orders to only participate at specific venues, such as only certain "dark pool" ATS. Also, they may(or not) want their orders to be included in the broker's internalization stream.
Special Parameters
Visibility, Benchmark price
Pair trading
Pair trading involves buying one asset whilst simultaneously selling another. Therefore, this is a market-neutral strategy; the risks from each asset should hedge or offset each other.
There are two main types of pair trading, namely statistical arbitrage, and risk(or merger) arbitrage. Statistical arbitrage is based on relative asset valuations whilst risk arbitrage is more equity specific, it revolves around the probability of a merge happening. A common approach for statistical arbitrage is based on the expectation of mean-reversion. This assumes that the spread or ratio for the prices of two highly correlated assets will generally oscillate around its mean.
For those interested in this topic the book "Pairs Trading: Quantitative Method and Analysis" by Ganapathy Vidyamurthy(2004) is a good place to start.
Basic Mechanism
The simplest example of a pair trade is based on the spread between two asset prices. When the difference in prices between the two assets exceeds a specified threshold, trading is activated. This may be based on the mid prices, or the bid price for the asset to be bought and the offer price for the asset to be sold, or even all possible combinations.
Common variations
Risk arbitrage pairs can generally use the same approach as the statistical arbitrage ones. The trading strategy is to sell the shares of the bidding company and buy those of the target company. If the merger happens, the position may then be unwound. The profit is from the difference in spreads; generally, the spread is wider before the merger then tightens as the deal reaches completion. Some versions may also support a cash adjustment, which may be applied before the ratio, or spread is calculated.
Special Parameters
Spread/Ration, Order identifier, Legging, Volume limit
5.7 Other trading algorithms
Multi-leg
Volatility-driven
Gamma Weighted Average Price(GWAP)
5.8 Summary
- A trading algorithm is simply a set of instructions used to execute an order.
- Trading algorithms may be broadly categorized into three main groups based on the target objectives. These are impact-driven,cost-driven, or opportunistic.
- Impact-driven algorithms seek to minimize the overall market impact costs, usually by splitting larger orders into smaller child orders.
- TWAP(time-weighted average price)is often driven by a time-based schedule.
- VWAP(volume-weighted average price) often uses historical volume as a guide.
- Percent of volume(POV)algorithms "go along” with the market volume.
- Minimal impact algorithms use ATSs dark pool to reduce signaling risk.
- Cost-driven algorithms aim to reduce the overall trading costs.
- Implementation shortfall(IS)seeks to achieve a balance between cost and risk.
- Adaptive shortfall algorithms extend this adapting to the market price(or liquidity)
- Market-on-close algorithms target the future closing price.
- Opportunistic algorithms strive to take the best advantage of favorable market conditions.
- Price inline algorithms are price-sensitive variants of impact-driven algorithms.
- Liquidity driven algorithms are an evolution of simpler rule-based order routing
- Pair trading is a market-neutral strategy driven by a favorable spread or ratio.
- Most of these algorithms will work across assets as the unique features of some asset classes mean that completely new types may also be needed. For instance:
- Multi-leg trading for bonds and derivatives
- Algorithms driven by factors such as interest rates or volatility
- Handling fungible assets, so specification-based trading rather than explicit assets
- Targeting new more appropriate benchmarks