HAL is statistical model design to detect specific market configuration and exploit them by assuming that price will be the adjustment factor. This model is only available for FX markets depending of the accuracy and availability of the data. It returns a signal output which is translated into a sell or a buy order over a defined maturity of 1 day.

I concentrate my research on a day to day algorithm trading. I’m more comfortable with this framework for a lot of reasons such as slippage impact, transaction costs or trading systems implementation.

A short summary regarding the model, I developed the alpha model during the summer 2010 and I implemented the first backtest over the last 10 years  ( to validate the model approach). The conclusion was very encouraging and I decided to set-up the system by creating a tool to trade on the market. It was the V2.

From September to December, I’ve replicated HAL on the market via CFDs with an initial investment of 450$. It was very volatile.

Then, I decided to think about a Risk model to avoid other November month, and this volatility too high . The chart 2 shows the NAV/share backtest (without Risk model) behaviour over year 2010 with an initial amount of 1500$.

Finally, some explanation regarding the model upgrade I used: any new major modification which needs to run a backtest is captured by the first figure (1.x). The second figure is linked to the calibration of the indicators and their threshold (X.1).

Every release should follow these steps (note that each step has is own process):

  1. Collect the data (assess accuracy and quality)
  2. Validation of the model (define a framework where indicators and target are defined)
  3. Market Implementation (set-up the tool and track any possible coding error)
%d bloggers like this: