The Flyer FOREX Fund (FFF) began trading real money in 2010. The only student-run fund in the country, it currently has 20 students and an 18-member advisory board. It generates alpha by trading the four-hour euro and Australian dollar spot markets using a non-discretionary algorithmic model. The model, which is not coded and manually calculated, uses trend indicators such as exponential moving averages (EMA), momentum, volatility and the girth indicator (see “Adding girth to your profits,” December 2008). The techniques are technical. The students chase the crowd, using trend volatility as the primary trend indicator and timing techniques for identifying the sweet spot for profit.
Since inception, the FFF has produced a 30.73% gross return through Sept. 28. Capital under management has grown by 312%. While actively trading, the fund has had a maximum book leverage of 3.3:1 and a minimum book leverage of 2.2:1. We began trading with approximately $2,000 and we currently have about $12,000 under management.
Because of the ever-changing market environment, we are updating and innovating our model constantly to optimize it. Since 2010, we have developed 10 models in addition to the original. Model development has overtaken our risk capital and, at maximum leverage, we are over-leveraged slightly. With a maximum target risk leverage of 2:1, we would need to increase our risk capital to approximately $25,000. Actual results are shown in “Fund results and targets” (right).
While the FFF is run by students out of the Hanley Trading Center, University of Dayton School of Business Administration, the fund is owned by Professor Leslie K. McNew (leverage is beyond of the scope of money managed by the university). The primary sources of fund capital are McNew, student dues and dues from advisory board members.
Since 2010, the Fund has had a mandate to trade the spot euro. A new model began trading in August 2012: V3.B. This new model is composed of what’s referred to as the Standard Style Model and the Global Euro Chaos Hedge.
The chaos of the European credit crisis, beginning in 2008 with the collapse of Iceland’s banking system and the U.S. credit crisis, has resulted in divergent euro market scenarios: Consolidation or extremely whippy price action. These situations have created speculative opportunities (see “Profiting from chaos,” below). Initially, to offset negative returns in euro positions, a model was added to trade the Australian dollar in the spring of 2012.
The Standard Style Model trade takes a position in euro and Australian dollar spot markets. We trade a €10,000 spot position with a margin of $300, and an A$13,000 spot position with a margin of $312. All trades are manually executed every four hours, 24 hours a day, Sunday through Friday.
Trade entry is based off the EMA signals. When the fast EMA crosses over the slow EMA, we enter a long position and vice versa. We spent considerable time backtesting optimal EMA combinations. The EMAs for the euro Standard Style Model are 10 and 20, and those for the Australian dollar Standard Style Model are 6 and 12. For example, in the euro, if the 10-period EMA is greater than the 20-period EMA, our position is long the euro at the next open of the four-hour candlestick. We can see from “Diversified risk” (below) that the Australian dollar position (red) usually acts as a hedge against the euro position.
Our trade exit depends on an indicator known as girth, which simply is the linear distance between the fast and slow EMAs. When the EMAs cross and we are entering a position, girth is at or close to zero. The stronger the trend, the wider the girth, thus the more profitable the trade. Girth is zero at the start of the trend and returns to zero at the end of the trend.
We segregate girth as upward or downward (see “Categorizing girth,” below). Upward girth refers to when the trend still is building; downward girth refers to when the trend is fading. We have found that exiting on upward girth is most profitable for the euro positions and exiting on downward girth is most consistently profitable for the Australian dollar positions.
Additionally, girth may be both static and bifurcated. Static is a single girth for both the long and short positions. A bifurcated girth employs a different exit strategy for the long and short positions. We have successfully backtested and have implemented a bifurcated euro girth of 35 for long trades and 45 for short trades, but employ a static girth of 20 for all Australian dollar positions as of Sept. 26.
One interesting feature of our new model has been the proof of the link between daily historical price volatility and euro girth readings. Higher daily historical volatility of the underlying usually produces a much higher girth number for a more profitable trade exit. From 2007 to the present, we have found that the historical euro 20-day rolling volatility average is 9.47%. We can optimize our girth exit by increasing the exit indicator when volatility is greater than the average. Our average static girth for our euro standard style model is 28. In this current period, the euro spot daily historical volatility is greater than 9.47%, so we widened out our girth exit to 35 for a long and 45 for a short position. As of yet, we have not tested this relationship for the Australian dollar.
Both of the standard style models have individual trade stop-losses. Through backtesting, we have derived the optimal stop loss of 300 pips for the euro positions and 100 pips for the Australian dollar positions. Because the manual execution of our trading position could result in the possibility of operation errors, we also utilize a global stop loss. We have a global stop of 8% in the case of an extreme loss situation and the market adversely moves against us within a four-hour period.
The chaos hedge
The Global Euro Chaos Hedge model employs a correlation trigger based on the daily correlation trigger of the euro against the Australian dollar. This hedge is a global optimization mechanism to raise profitability for the euro and Australian dollar Standard Style Models because of the European credit crisis. It is based on a backtested correlation trigger of 63%.
We have found that when euro/Australian dollar correlation drops below 63%, the euro market produces non-trending volatility while the Australian dollar market is more trending (see “Rolling correlation,” below). Because our Standard Style Models profit off directional volatility trends, when correlation drops below 63%, we want more of our Standard Style Model exposed to the Australian dollar market. The 63% correlation trigger activates our Chaos Hedge, which doubles our Australian dollar trading position at trade entry on the open of the four-hour candlestick; the additional Australian dollar position is held until girth indicates an exit.
The backtesting results of this new model (V3.B) that combines the correlation trigger with the Standard Style Model look promising. Results are shown in “Chaos filter” (below).
FFF has three major risks: Market, credit and operational risk. Regarding market risk, we have built a two-factor parametric portfolio value-at-risk (VaR) model to illustrate the actual risk of the fund. We have no minimum VaR as every year we have periods where trading is halted; thus, VaR drops to 0. The actual VaR of our fund is shown below.
|Year||Avg. daily (95%)||Max|
The FFF has credit risk exposure to both JPMorgan Chase and FXCM. JPMorgan holds approximately $6,000 of resting capital and FXCM holds about $6,000 of risk capital. If either one of these institutions were to fail, we still would be able to manage our program, although leverage would increase.
We still have a long way to go in controlling operational risk. Operational errors for the student-managed fund remain numerous (see “True cost of learning: model implementation,” September 2011).
The FFF uses a benchmark as a point of reference to measure performance. This benchmark is the return of the active model as if it were traded continuously and executed automatically at the beginning of the four-hour candlestick. As the model has evolved, performance with respect to the benchmark has improved considerably.
There are three factors that can cause a difference between actual and benchmark returns. These are slippage, operational errors and downtime (surpasses the other two factors). The FFF has had downtime in the following areas: Stopping trading to upgrade a model, university holidays when students refrain from trading and the faculty manager halting trading because of excessive operational errors. In only one instance did the FFF team outperform the benchmark: Summer 2011 when the faculty manager guided trading to test a market theory that resulted in a model upgrade.
The alpha of the FFF is in its model and the students who successfully build and manage the model, its upgrades and its risk. During the spring term, an algorithmic modeling class will teach advanced Excel and Visual Basic for Application skills and will work to provide a new upgrade to the existing model. This class is limited to FFF members and is invitation only.
This past summer, an internship program produced a complete quantitative overhaul of the existing model (to find validation errors), built a new upgrade to the model, built the trading benchmark, built the-two factor parametric VaR model, built the volatility/girth overlay, produced trading returns in an acceptable format and validated all work processes with an advisory board member.
Research support for this article was contributed by Christine E. Muench (’13), William M. Schaid (’13), Christopher J. Waldock (’14), Samuel C. Fister (’14) and Christopher E. Bell (’15).
Leslie K. McNew is a clinical professor of finance and director of the Hanley Trading Center, Department of Economics and Finance at the University of Dayton Business School. E-mail her at firstname.lastname@example.org.