Econophysics in Action.
We are a group of passionate scientists, traders, and developers dedicated to democratizing algotrading for small investors and independent trading firms. We believe that applying the scientific method and a natural science worldview to finance gives us an edge and consistency of results. Our algos are mostly based on the Online Portfolio Selection (OLPS) approach and we created our own proprietary OLPS that outperform most OLPS algos in the available literature.
Online Portfolio Selection: A Survey
Our algos pass very strict performance and stability tests. Our benchmarks are always the best performing ETF on the market (e.g. QQQ for stocks) and our aim is to show returns that are consistently better than the benchmarks. We developed what we consider a unique approach to trading and a methodical process to fintech R&D.
Quantonomy Trading Manifesto:
1) Markets are predictable, the efficient market hypothesis (EMH) is either wrong in general or on short time scales ranging from minutes to several days. Many inefficiencies in the market can be exploited.
2) Trading successfully means to not simply react to the market, but predict its behavior.
3) The majority ,if not all, of the methods based on a single asset time series that attempt to identify entry and exit points are reactive and not predictive. At best, they may succeed in identifying low and high turning points in the time series, but they will always be late because of problems like noise filtering. As a result, these methods lack predictive power. This also applies to pair trading.
4) Understanding a related group of assets as a whole is a much more powerful trading strategy. This approach aims to capture changes of multiple assets relative to each other. It is possible to find simple predictive metrics of performance that allow ordered ranking of the assets based on the predictive metrics. The metrics then can be used to predict the important future behavior of the assets as a whole , e.g. relative returns in the near future. Demonstrating statistically that the predictive measure can indeed predict the properties of the assets in time is fundamental.
5) Focusing on the behavior of the group instead of single assets results in trade-off between capturing the price action of a single asset and understanding how a group of assets organizes as a whole. This means the exact return, or sometimes even the direction, of an asset cannot be predicted, but winners and losers relative to the group can.
6) Always start from the simplest and intuitive metrics and the relationship between asset properties and the quantity that has to be optimized (cumulative returns, Sharpe, Sortino, and similar). The input data used is price first and volume second. Complexity should be added with caution. For example, algorithms with more than 2 parameters are not ideal. Simple ideas from Machine Learning are fine, however black box systems like intricate, multi-layer Deep Learning algorithms are not.
7) Make the strategy adaptive to the ever-changing market conditions. Use walkforwards methods vs static backtesting.
8) Continuously monitor and characterize the trading strategy over time to identify possible problems, inefficiencies, and signs of alpha-decay. Quickly correct the problems and improve the strategy over time after collecting enough data to make informed decisions.
9) Make several strategies compete with each other by "optimizing" between them with the use of various methods.