Platform

 

PLATFORM

Our trading platform is capable of generating quantitative knowledge and making the most of the information contained in the market data; it is made up of 4 modules:

MARKET CLOCK

Focused on Intraday, aims to providing a description of the
most probable market behaviour
for the day ahead.

PATTERN RECOGNITION

CURVE VIEW

Curve View: increases the accuracy of relative value, exploiting
“deep” structures hidden in correlated assets.

AUTO ASSOCIATIVE NETWORKS

FORECAST

Information about “asset universe”
and historical asset behaviour
delivers a bias on its performance.

RECURRENT NETWORK

CROSS ASSET

Provides information about market
drivers, highlighting investment
opportunities and systemic risks.

VARIATIONAL AUTOENCODER, RBM

FIND OUT MORE

Market Clock

“History doesn’t repeat itself, but it does rhyme.”  Mark Twain

Market Clock is:
1. An ensemble of pattern recognition algorithms for forecasting -> technical analysis
2. A recurrent network -> short term forecasting
3. A tool to summarize information contained in correlated assets -> cross asset evaluation
4. A system to monitor the performance of different algorithms in the past and online -> backtesting
5. Robot trading  -> online trading …$$$$

1. PATTERN RECOGNITION: FROM TECHNICAL ANALYSIS TO MARKET CLOCK

TECHNICAL ANALYSIS 

• Known pattern recognition (finite selection of pattern)
• Trader identifies formation and behaves based on past market behaviour

Candelstick scanner

MARKET CLOCK

• Known and unknown pattern recognition
• Algorithm identifies most probable «trajectory of the market»
• Provide a probability distribution for an end of day return.

Market clock output

OUR APPROACH TO PATTERN RECOGNITION

  • We observe the time series, properly normalized in a trajectory.
  • We assume that there is a finite set of typical or “archetypal” trajectories (“market states”) that represent intraday market behavior.
  • These trajectories are created by the interactions between different players and different speed-moving capital
  • We design algorithms that can predict the “most probable” destination, looking at the partial trajectory at a subsequent point in time.

ONLINE FORECASTING

In the framework we have created, the challenge is to:

Find a good representation of the data set in terms of «typical» trajectories.

⇒ Unsupervised methods for clustering

Find a performing algorithm to «learn» how to connect the online observation with the «most probable» trajectory for the day.

⇒ Supervised methods for forecasting (NN for classification, svm

MARKET CLOCK ALGORITHMS FOR PATTERN RECOGNITION

In Market Clock, we currently use 5 different algorithms:

  • a prior K-means clustering
  • a posterior K-means clustering (transition matrix)
  • a multilevel SVM classifier
  • a MLP for classification
  • a Deep Tapped Delay Line with enhanced clustering
Algorithms performances are continuously evaluated
New algorithms are added if backtesting yields good results or vice-versa.

2. MARKET CLOCK ALGORITHMS: RECURRENT NEURAL NETWORK

In market clock we also propose a recurrent network to improve prediction.

• Generalization of Markov Models
• As we have sequences of “words”
we can use the tools used for Natural Language Processing
• Generating a time series is like generating a text
• The recurrent neural networks can “forget”
• The output is just a shifted version of the input
• We train the network to interpolate the “unknown” values in the input vector
• The network extracts the probabilities for each of the codes for the next steps

3. CROSS ASSET EVALUATION 

The algorithm performance can be improved significantly using information from correlated assets.
In the module, we synthesize information coming from similar and correlated markets using machine and deep learning algorithms in real time.

The fair value coming from correlated asset is summarized by the two lines (purple and yellow) evolving in time.

Usd chf, 11 Jan – 19

The same framework allows us to enlight the main drivers of the day and to extrapolate the «low frequency» trend. In fig1 and 2 the cross assets and currencies drivers for the day, with the decomposed trend in the bottom box (white line).

Usd chf, 14 – jan -19

4. HISTORICAL PERFORMANCES: MODULAR BACKTESTING

In Market Clock we use a modular approach to backtesting. This approach allows us:

1. to reduce the number of run of the algorithms;
2. to optimize separately the algorithm forecast and the trading strategies (algorithm ensemble or scoring)

The datasets are partitioned usingrollingwindows

The models are trained and simulated over a gridofparameters

For each model (and for more than one model at a time) different threshold strategies are used to generate trading signals

Using all the out puts of the proceeding steps a market simulator evaluates the PL of the pair algorithm plus strategy

The backtesting results are continuously updated in order to evaluate the relative algorithms’ performance and to verify the online performance of the algorithm with the backtesting results.

Usd chf, 11 Jan – 19

5. ROBOT TRADING

In the platform, you can activate a «robot trading» status, after proper configuration of the stop/target strategy and signal level. We use IB api to TWS.

For more information fill out the form!