Dataset Name: tesseract

CloudQuant long-short backtest results show the following:

The Tesseract Signal is a set of 150-200 symbols using unsupervised & supervised learning methods to design a portfolio with the highest probability of outperforming the S&P 500 index. Symbols are rated from 0 to 1 on their likelihood to out or underperform the SP500.

CloudQuant Finds Tesseract Data Provides Significant Alpha

Alpha in The Data Set

The returns for Experiment #3 : Long the top 10%, Short the bottom 10%, Hold for 10 days.

CloudQuant has found that going long the top 10% and short the bottom 10% of Tesseract Machine Learning Signals (Tesseract Signals) returns an average of 12.70% per year after transaction costs for two and half years.

We have found that over 80.0% of the total return is pure alpha not explained by traditional market, size momentum and value factors. The results are significant to the 99th percent level. The realized signal returns show no significant correlations to any known "smart-beta" factors. In combination with other factors, the Tesseract signal is beneficial to the construction of most fundamental hedged and long-short portfolios.

White Paper Research Conclusions

CloudQuant finds the Tesseract Signal identifies both long and short investment signals that produce statistically significant investment return (alpha) at a greater than 99.9% (p-value < 0.001) level of confidence from 2018-2020 on the Mariner strategy backtesting platform.

The top portfolio was Experiment 3 (Top 10% Long, Bottom 10% Short, Holding 10 Days) which returned an average of 12.70% per annum , and had yearly Sharpe Ratios of 3.9, 5.6, 1.3 for each year of 2018, 2019, 2020, a beta estimated to be 0.014 (p-value < 0.05) and an alpha of 15.32% (p-value < 0.0005) with 99.9% confidence.

The higher the score is, the stronger the alpha signal will be. Similar conclusion also stands for short sided signals. By comparing long-sided and short-sided portfolio performance alphas, there's evidence showing stronger signals in bottom quantiles than top quantiles.

These results are remarkable given univariate nature of strategy and the simplicity of the equal-weight, dollar-neutral portfolio construction process. One-sided-signal portfolios (Experiment 7) showed obvious exposure to beta, resulting in lower returns and higher volatility. Two-sided-signal portfolios (Experiment 1-6) showed little exposure to beta, and a strong alpha each year.

The Tesseract Signal is highly unique and distinct from other known risk factors. Return decomposition analysis and correlation analysis comparing Tesseract Signal portfolios to known "smart-beta" risk factors and commodities shows that over 80% of the dollar neutral strategy return is comprised of pure alpha (idiosyncratic). The portfolios having negative factor exposure to XLV, USO, and some positive exposure to XLK, XLE and XLU.

We believe that in combination with other alpha signals much higher returns and Sharpe Ratios are achievable. More possibilities can be explored by looking into the auxiliary features inside the database to extend the holding period of the strategy, and to lengthen the lifespan of the alpha.

The Tesseract Signal

The Tesseract signal is based on a novel combination of unsupervised and supervised learning methods. A multidimensional scaling procedure is used to measure the underlying factors driving returns for equities in the S&P 500.

Using a more precise measure of fundamental characteristics such as quality and momentum, a set of tree-based ensemble models examine how these factors relate to economic and market conditions (for example, volatility, sentiment, and fund flows) to design a portfolio with the highest probability of outperforming the S&P 500 index.

Click here to "Schedule a Demo" or Email Sales@cloudquant.com to obtain the in depth research which includes the following:

• CloudQuant White Paper - Evaluation of Tesseract Dataset
• A demonstration of the CloudQuant algorithms
• Source Code and free access to CloudQuant Mariner to replicate the study results and/or evolve the strategies to your requirements.

Data Contained in this Dataset

Important Dataset Notes