The TrueRisk Labs Advantage

Our team has spent years developing our AI models which are able to aggregate thousands of financial and quantitative metrics along with Big Data, and rank and sort that information faster and more consistently than human analysts or quantitative traders.

We deliver actionable insight. Our deep-learning algorithms are self- adaptive and uncover new non-linear patterns before they have a market impact. Our AI models autonomously change their outputs based on the changing inputs.

The TrueRisk Labs Quantamental Dataset is real-time, alpha-generating analysis of Big Data.

TRUERISK LABS

TRUERISK LABS – ARTIFICIAL INTELLIGENCE

Aggregated Data using AI models that are Self- Learning and Autonomous

Adaptive
QUANTITATIVE TRADERS

QUANTITATIVE TRADERS

Decisions Based on Statistical Mathematics and Arbitrage

Systematic
FUNDAMENTAL INVESTORS

FUNDAMENTAL INVESTORS

Decisions Based on Company & Macro Fundamentals

Judgement-based

AI Models Employed

We use ensemble meta-algorithms for primarily reducing bias and variance in supervised learning, and to convert a family of moderately-strong learners to a strong and robust one.

Our choice of method for each incoming dataset is dependent upon the structure of the raw data and proceeds through a system of checks and balances to ensure accuracy. We then employ deep-learning algorithms to layer generative models into feedback loops which are able to learn and adapt to changing conditions.

Our system was developed in collaboration with experience portfolio managers and traders from the worldʼs top firms. We understand risk and we guide our AI to produce the best risk-adjusted outputs.

LINEAR AI MODELS

  • Logistic
    Regression (LR)
  • Linear Discriminant
    Analysis (LDA)
  • Bayesian
    Regression (BR)

NON-LINEAR AI MODELS

  • K-Nearest
    Neighbors (KNN)
  • Classification and
    Regression Trees
  • Gaussian Naive
    Bayes (NB)
  • Support Vector Machines (SVM)

DEEP LEARNING NON-LINEAR AI MODELS

  • Artificial Neural Networks with Backpropagation for supervised learning
  • Restricted Boltzman Machines (unsupervised learning)
  • Recurrent Neural Networks (RNN) for time series data
  • Long Short Term Memory (LSTM) Neural Networks – A Class of RNNʼs specifically suited for time series data with feedback loops (like stock markets)

Good Inputs = Good Outputs

The TRL Quantamental Dataset is designed from the ground-up to output the best risk-adjusted ratios. Our AI models are guided by the way we structure incoming data and the specific model employed for each dataset with anchoring to predicted volatility as well as price movement.

The collaborative experience of our risk management and quant teams is able to provide the critical insight to handle incoming data in-line with principles of sound fundamental and quantitative analysis.

  • Price/volume data across multiple timeframes from domestic and international securities
  • Quarterly released fundamental metrics and self-calculated ratios from every company in our coverage.
  • Price/volume and technical studies of global currency, commodity, and bond markets.
  • Quantitative correlations and patterns between all data points and feedback loops of our model outputs
  • Sell-side analyst forecasts of price, revenue and earnings on every security in our coverage
  • Technical analysis indicators across multiple timeframes on every security
  • Money-flows from domestic mutual funds and international equity exchanges.

OUT-OF-SAMPLE DAILY WALK FORWARD TESTING ACCURACY 2013-2018 ON WILSHIRE 5000 EQUITIES

58% Correct analysis across 5 years

3,492 Stocks Analyzed

Absolutely correct predictions are defined by stocks which move beyond our long/short price targets within the boundaries of expected volatility and time frame.

TRL's proprietary Ai models and deep-learning algorithms allow us to provide unique insights into expected stock price and volatility movements in 3-month, 6-month, and 1-year time horizons.

Our AI models began learning in 2011 and out-out-sample testing began in 2013. There has been no fitting of the data.