AutoMetis is pleased to unveil its proprietary automated trading strategy "Athena": a systematic method to monetize fundamental technical properties of FX markets.

Athena was conceived in 2015, and had a prototype launched into live trading in the FX market later that year. It has since received two major overhauls: one in late 2017 relating to an improvement in its execution algorithm resulting in savings in transaction costs, and one in early 2020 relating to its trade order diversification mechanism and to a recalibration of its risk framework to bring it in line with the market standard.

We present below in detail the underlying concept, the backtested performace, our live track record, and the risk analysis of our stress-testing simulations.

• Absolute return strategy constructed to achieve a high-certainty return target
• Medium frequency, fully automated trading over major USD crosses
• Uncorrelated with equity market; scalable and flexible risk profile

Athena exploits the existence of fat tails in the distribution of FX price changes in the medium term — the empirically-supported proposition that large market moves occur more frequently than normally.

At the same time, the academic literature finds in the FX market the least evidence for the existence of mean reversion, further fuelling the expectation that deviations from the mean tend to be large and lasting, while the academically established mean reversion for volatility suggests that periods of a tightly bound price range have a finite duration.

Reflecting these ideas, Athena remains directionally agnostic, selectively overtrading but avoiding short-term noise, to maintain alignment with the market in order to eventually capture a move sufficiently large to recover accumulated losses and yield net profits.

Athena also presents a valuable advantage for risk management: the accumulation of short term losses is slow and gradual, ensuring an inviolable risk framework and allowing for comfortable monitoring and assessment, while also offering significant immunity to violent market shocks, which are likely to enhance rather than threaten performance.

In offering high certainty in its long term return expectations, Athena is faced with a particular, relatively uncommon, kind of risk: that of a moderate (above noise levels, but not high enough) and protracted sideways volatility. The statistical quantification of the duration and depth of resulting drawdowns is presented in the Risk Analysis section.

The algorithm determines a key level around which it trades, subject to an internally determined noise threshold.

When the FX Price fully traverses the noise channel, the algorithm fully reverses the trade's direction. Each such reversal realises losses, whose magnitude depends on the noise threshold and on transaction costs (bid/ask spread and slippage).

Unless such a reversal occurs, the particular trading position is maintained until an internally determined take-profit level is met. The take-profit level ensures recovery of the cumulative losses of that trading cycle, as well as a variable amount of net profit.

The strategy benefits from diversification across numerous such channels being placed at different key levels, as well as from the natural diversification across the different currency pairs traded. Furthermore, two separate sets of inputs into the major determinants of the algorithm are deployed in simultaneous implementations for each currency pair, to avoid overfitting bias.

• Could the strategy be applied to asset classes other than FX?

  Perhaps. As discussed in the Concept section, the FX market evidences the smallest adherence to mean reverting behaviour and is therefore the domain of choice for this strategy.

  Additionally, some features of the FX market make it comparatively attractive: deep liquidity which allows for scalability and small execution slippage; 24/5 trading continuity which minimises gap risk; lower spreads and lower overnight financing costs than most other markets.

  Consequently, we have chosen to limit our research and strategy implementation to the FX market for now.

• Could the strategy be run over additional currency pairs?

  Yes, although we do not expect to be able to extract sufficient diversification benefits in order to significantly improve the overall risk-adjusted performance.

  While the strategy is not high frequency, transaction costs are a significant hurdle to clear for any systematic strategy. The particular currency pairs are among the most liquid and cheap to trade, while at the same time each of these USD crosses has a macroeconomic link that supports the expectation of a sufficiently high long term volatility.

• What is the minimum capital required to run this strategy?

  For the risk/return profile presented above (i.e. 9% p.a. return / 20% volatility / 36% expected max drawdown) the required risk capital base is $250,000.

  The strategy can be run on the basis of less risk capital, but the risk/return profile will be negatively affected — this is due to the fixed minimum size of individual trades imposed by the broker.

• Is the strategy fully scalable upwards?

  Yes. Our execution algorithm as it stands could support a 100-fold increase to a $25m capital base with no additional changes.

  Beyond that, it would probably be wise to overlay an order submission mechanism to spread out the trading. The strategy is not high frequency and therefore has room for additional timing optimisation, which is not currently relied upon.

• Does the strategy require leverage? What are the risk and capital implications?

  Trading in FX is typically done using some kind of a derivative instrument, such as a Forex contract, a CFD, or a Spread Bet. These instruments are all traded on margin, meaning that the deposit capital required by the broker is a fraction of the net notional value of the trading position — or equivalently, the broker funds a trading position which is a multiple of the deposit capital required, implying leverage. For major FX pairs, depending on the client categorisation by the broker, the required margins are relatively small; a professional client could trade at 0.3% (over 300x leverage).

  The risk consequences are not inherent to the leverage available but to the Value at Risk at any time, which depends on the notional exposure in combination with the allowed loss.

  Indicatively, during our Live Period, our peak aggregate notional exposure reached just 6x Risk Capital, while the largest loss on a single trade was -0.83%. This mild leverage implies less than 2% of our Risk Capital being locked up for margin requirements — a small enough amount to be negligible for the sake of simplicity in the presentation of the strategy.

• How is the strategy different from a long options straddle strategy?

  Access, liquidity, cost, and horizon:

  The strategy trades in the spot FX market, so is accessible by virtually anyone, it does not depend on a particular market maker, and it is not exposed to liquidity dry-up risk.

  The transaction costs are fixed and small, compared to buying two (likely overpriced) at-the-money options.

  Finally, the strategy achieves a similar payout profile but without the constraint of a finite horizon for the realisation of its target.

• Does the strategy's success imply that the FX market is inefficient?

  No. The performance does not necessarily represent excess or abnormal return, nor is there any evidence that the market does not properly and quickly discount all available information.

  The return achievable can be viewed as market compensation for the strategy's variance risk, as well as for the strategy's negative skewness.

• Are you planning improvements to the strategy in the near future?

  We believe the strategy currently reflects the underlying concept well, and we have already made the necessary improvements to our execution algorithm.

  We are reluctant to tweak the inner working of the strategy, as it would have questionable incremental value and give rise to potential overfitting biases.

  Instead, we are now focused on developing a framework for the management of the strategy as a whole: predicting when the strategy tends to perform well and when not, and systematising a weight allocation to it.

• Aren't you worried you are disclosing too much detail about the strategy?

  On the contrary, we prefer to adopt a transparent approach in order to build confidence and invite further discussion and contributions.

  The strategy is not purporting to rely upon some holy grail discovery or secret sauce; the fundamental concepts have long been explored in academia. Instead, our key value is our work on the execution of the idea and the construction of the implementing framework.

  We specifically aim to differentiate our approach from the all-too-common "black box" products for sale we see in the market, which as a rule have dubious underpinnings and rely on cosmetic manipulations to paint an unrealistic image of their potential.

• On what platforms can the strategy be run?

  We have been running the strategy on the MetaTrader 4 platform of our broker, IG Markets, using a partner VPS service.

  The strategy, currently coded in MT4's MQL language, could be re-coded to run on other platforms or link directly to a broker's API — all that's needed is a live and continuous feed for the underlying prices and a modest amount of historical data; everything else is built in, there is no reliance on external libraries, indicators, or databases.

• Could I send you some capital to run this strategy for me?

  No, sorry — AutoMetis is purely a software development company, and is not licensed to undertake regulated financial activities.

  We are exploring partnership options with relevant institutional entities, as well as with platforms offering copy trading capabilities.

  If you have your own trading account and would like assistance to code a systematic strategy, we could be able to help — please use the form below to contact us for further discussion.

Name

email

Message

Send