The Gateway to Algorithmic and Automated Trading

Mad-aunt strategies

Published in Automated Trader Magazine Issue 14 Q3 2009

Has the gap between forecasts and reality widened in recent years, and if so, should we adjust our use of forecasts? For Helen Sanders, accuracy is in the eye of the end-user.
Has the gap between forecasts and reality widened in recent years, and if so, should we adjust our use of forecasts? For Helen Sanders, accuracy is in the eye of the end-user.

Where are you now? Reading this article? No, this is not an invitation to Twitter your whereabouts to the world. Are you in a departure lounge? On a plane, or at a trade show, or waiting for a meeting to start? Reading online on a Friday afternoon? Wherever you are, there's a good chance that you're within, let's say, ten yards of a certain advertisement. Just as city dwellers can address their irrational exuberance with the thought that in a city, nobody is ever more than ten metres from a rat - so you, as a private individual with an interest in matters financial, are probably not more than … let's say ten metres from an advertisement reminding you that (allowing for regional variations) "past performance is no indication of future returns", and that "forward-looking statements" are not actually guaranteeing the predicted outcome.

As if we needed to know that. The past year has emphasised painfully that such statements amount to a fundamental - albeit obvious - market principle.

So. Knowing what we know now, have we stopped relying on forecasts for the future? When nine out of ten consultants tell us that fragmentation leads to consolidation, and that the "legacy" exchanges will be no match for the proverbial two guys in a conveniently located garage with a server and a significant market share of equity trading in Europe - do we pay them for that insight? Or invite them to show us where and when exactly they predicted the meltdown?

One of the curious phenomena of modern times is that forecasts - not just the big, official stuff, but the pundits and the whacko characters in red braces with a crystal ball showing on their laptop screens - keep on coming.

Forecasting is discredited; long live forecasting. And the other curious phenomenon is that we go on acting as if we believe (some of) them. Or at least - as if having a forecast is more useful than not having a forecast, even though we know that said forecast is as related to what's going to happen as … a certain bank's last annual report is related to what happened to it.
It's time to revisit forecasting. Over the past two years, it would seem reasonable to suggest that the "uncertainty gap" between forecasts and outcomes has widened. And we're not even talking here about the choices one has to make (or deliberately not make) when back-testing a new trading model. We're talking about our new knowledge of the [actuaries, please look away here] possibility that the impossible will
happen; of the probability that [mathematicians, this would be a good time to break for coffee] million-to-one chances can happen about one in ten times; and the likelihood that [linguistic purists - go! Just go!] bricks'n'mortar aren't as secure a bet as money in the bank, which itself isn't as safe as houses.

Even without breaking the language into rubble, it's a tenable hypothesis that the least likely outcome is the one specified in the forecast. And if that prompts you to think of a forecast that's reliably (mostly) accurate - okay, fair enough, but think of a few that aren't so reliable, and ask yourself whether there's anybody out there who's using them as a basis for trading decisions. Yes?

You will find this article very interesting

Data appetites

Let's analyse this. One of the essential elements of virtually any society is planning. We love to plan. From the farming calendar, government spending plans and company strategy, through to "to do" lists and how to fill endless school summer holidays - all of these require us to look to the future and make decisions. We know from experience that the school holidays will be warm and sunny, or cold and rainy, or a combination of the two, and we plan accordingly. But what happens when a mad aunt comes to stay unexpectedly, or the house floods, and our summer works out quite differently from the one we anticipated? From the point of the view of the financial markets, the crisis has been a mad aunt coming to stay. Unexpected, undesirable and a fundamental change to all the assumptions on which trading decisions were based.

Forecasting is essential for algorithmic traders, to try and anticipate how markets will respond to market events and seek opportunities. Accuracy is an element of that, but so is "reliable inaccuracy"; better to know that our favoured indicator is always around ten ticks up from the eventual outcome, and to "aim off", than to keep on hunting out accuracy. And this is key both to effective use of forecasts, and to a current potential error that has arisen in the wake of the financial crisis. Forecasts need data, but forecasts are not the same thing as data. As this implies, attempts to improve forecasts by "stuffing them full" of data are not guaranteed to succeed (that's our forecast, anyway). The key to using data in forecasting is to recognise it's limitations: data is only two out of three of the necessary components of a forecast.

Trading forecasts generally require three elements: historic data; information on current events, and assumptions. Looking at the first two of these, data vendors are making available ever-increasing amounts of data in more sophisticated ways. According to Yuriy Shterk, Vice President, Product Development at CQG, Clint Rhea, Chief Operating Officer, Need to Know News (NTKN) and David Knox, Managing Director at i-traders, customers are seeking more data points over a longer period. For example, a trader who previously looked for daily closing prices over 6 months may now be looking at multiple intra-day prices, even per tick pricing, for 12 or 24 months.
However, as Yuriy Shterk, CQG warns: "Some people know what they need the data for, many do not!" Having all the data is not the same as being able to use it erffectively for forecasting.

David Knox, i-traders, describes how his firm is helping traders to organize and disseminate data into something useable: "We provide daily research across eighteen markets at a highly detailed level, including the reasons for changes in price. What we are now trying to do is to take what we do in analysis and automation to provide a one-click outlook for each asset class, right down to individual equities. We are looking at our computer-readable news and tying it in with our research, to provide the technical data behind the news."

Commentators also agree that traders are seeking a broader spectrum of data. Clint Rhea, NTKN says: "Traders have been demanding more detail on events and the number of data points they need. They are also seeking data for a greater range of products, beyond the asset class they trade, such as implied volatility from options and credit default swaps, and correlations between asset classes."

Yuriy Shterk, CQG, suggests that seeking data on multiple asset classes is not simply an issue of refining forecasts but more importantly, the way that risk is identified. "More people are now looking at data across instruments whereas in the past they would look at trends in trading patterns for individual asset classes. This affects data requirements, but more interesting and more complex is the need to change the way that risk is monitored. Traders need to be smarter about recognizing signs in one asset class that could affect another."

The third element

The crisis has begged the question, are the assumptions on which forecasts are based valid today as they were a year ago? That is to say, will the given market/asset class move in an equivalent fashion to a piece of news as it has done in the past? The answer has been a resounding "No!" in two key respects: first, the number of extreme events skews all the risk models on which assumptions have been based in the past; secondly, with events such as bank and corporate failures dominating the headlines, other more routine news announcements have paled into insignificance.
As Clint Rhea, NTKN explains: "The past 12 months has seen the demand for, and effect of market information, go full circle. A year ago, data such as employment figures and interest-rate decisions were important announcements that would move markets. Then as bigger issues such as bank failures dominated the market, people knew that interest rates would go down or remain the same, and expected lower employment figures, so this information ceased to be so significant."

During periods of extreme market volatility, are forecasts any better than a stab in the dark or crystal-ball gazing? David Knox, i-traders explains that forecasts remain fundamental to traders. After all, without some sort of view about what a trading period may bring, where to you start in determining your trading strategy? Accurate forecasting requires constant refinement as markets evolve, in addition to the impact of extreme events. For example, anomalies between forecast and actual market activity can be important to build into future forecasts and decision making,

"Forward projections are a vital part of our research and analytics. We provide a daily report with a graphical representation of the previous day's forecast and the actual data, highlighting anomalies and explaining these if necessary. In an environment of high volatility, we look at several trading sessions and provide a summary of support and resistance levels."

Different data providers, as well as trading firms, will have different ways of creating forecasts. David Knox continues: "How we create the outlook depends on the market. We start by looking at the market profile and previous sessions. We then look at a longer term profile to bring in support and resistance levels, vacuums, overbought/oversold positions and then deliver the outlook from there. When a market is overbought or oversold, we use non-standard models to provide analysis. For example, a rally on a Friday may be grounds for caution on the Monday."

A related issue, as described by Clint Rhea, is that the extraordinary market events we have seen are themselves being built into assumptions so that forecasts better reflect reality: "Market phenomena such as quantitative easing have a significant impact on market assumptions and how these are projected. In our product, Lightning Bolt, we have added a series of questions in computer-readable format to take into account the effects of quantitative easing, improving traders' ability to forecast the impact of market events."

You will know what to do with forecasts now

The maverick opportunity

But is there a trading opportunity presented by routinely inaccurate forecasts? It is probably realistic to think that even though different models apply, the majority of market participants are basing their forecasts on a reasonably standard set of assumptions, and using comparable historic data. The maverick could seek opportunity by assuming that the forecast will differ by at least a certain margin from reality, and routinely build this into a trading strategy. This requires a strong risk model, however, to assess the impact of possible anomalies either above or below the forecast.

After all, the crisis has illustrated that a vital purpose of the historic data and assumptions used for forecasting is not simply to predict market events (which may be fruitless if, for example, a bank or major company collapses the following day) but to feed into risk models. A risk model analysing various outcomes, which traders can then use to make decisions based on acceptable levels of risk, is more valuable than a simple forecast, whatever the risk appetite.

Adam Sussman, Director of Research, TABB Group, takes the example of an equity risk model which looks at correlations between equity prices. He explains: "When companies implement quantitative trading strategies, one of the key elements is the equity risk model. These look at various relationships between stock prices over various periods of time: for example, the correlation between stock prices over a number of years or just a few days. Depending on the length of time taken, a certain number of data points are included for each day, from an end-of-day price to a defined frequency during the day, or even every trade."

Sussman continues: "There are two types of consumers of this risk data: the first is the trading strategist, and the second is the risk manager. As markets shifted from a relative period of calm to record levels of volatility, these consumers started to look at these risk models differently, particularly assessing how responsive they are to the new market environment and how realistic historic data is in creating forward-looking projections. Traders are also running risk calculations more frequently, not just daily but regularly intra-day. This changes the requirements traders have of their risk management providers to support this additional reporting burden."

There is has been a long standing debate between academics and providers about the approach that should be taken to producing these equity risk models. The fundamental approach makes assumptions about the relationship between equity prices, such as whether a sector is cyclical, seasonal or consumer-sustainable. The other approach is statistical, which makes no sector assumptions. This model takes all data and goes through a procedural process to identify correlations. This debate is now dead: traders are seeking both, and want to compare the results.
With traders having access to massive volumes of data, and the ability to create or buy in forecasts based on their own, or third party assumptions, this year should have proved a valuable lesson that data in itself is worthless unless put to a particular purpose. Markets do not always behave as people expect, or as they have done in the past, but this is not to underestimate the importance of a forecast. It is not necessarily an end in itself, but is vital to feed into a risk model which projects different outcomes depending on varying levels of anomaly. Traders can then refine their trading strategies in accordance with their risk appetite.

As Clint Rhea, NTKN indicates: "A year later, as markets start to achieve some sort of equilibrium, regular market announcements are becoming more important again."

The markets are returning to something recognizable, if not normality in the "old" sense, but the tools and analytics which traders have put in place, which rely on the process of forecasting to create risk models, will be as valuable going forward as they have been during the crisis - for a given value of "valuable". Use forecasts, but don't use them to tell you what's going to happen next.