- Backed by a math library SuanShu, which allows you to prototype mathematical/quantitative models very quickly
- strategy optimization
- Read historical data from various sources such as Bloomberg, CompuStat, Gain Capital, Yahoo.
Unified Mathematical Analysis and Modeling Library
AlgoQuant has a large collection of mathematical models, many from top academic journal publications, which a trader can use as building blocks to create his own trading models. For example, to build a mean reversion model, the trader can combine (D’Aspremont, 2011)to construct a maximally mean reverting portfolio of two assets and then trade the pair using (Elliott, van der Hoek, & Malcolm, 2005)pair trading strategy. Both mathematical modules are available in the library. In addition, AlgoQuant has a large number of algorithms that the trader can apply. For example, in the area of portfolio optimization, AlgoQuant covers Quadratic Programming(Markowitz or Modern Portfolio Theory), Second Order Conic Programmingas well as Differential Evolution. In other words, using AlgoQuant, the trader does not need external math software like Excel, R, or MATLAB, and there are many readily available modules and algorithms to use. He can quickly build up very complicated mathematical strategies by combining together the components from the library.
NM FinTech Inc. has a unique way of doing backtest. In general, backtest using historical data alone is not a sufficient justification of a quantitative trading strategy. Historical backtest tells only how much money we could have made in the past. What we really want to know is how much money we may make in the future. The future P&L is a random variable. Therefore, in addition to reporting the historical P&L, a proper backtest should, when applicable and feasible, report also the probability function of the future P&L.
In general, optimizing strategy parameters is a very difficult optimization problem due to, e.g., curse of dimensionality, discrete or integral variables, trading constraints. AlgoQuant comes with a suite of professional optimization tools, e.g.,LP, SDP, SOCP, QP, DEOptim, simulated annealing, genetic algorithm, to do strategy and portfolio optimization.
Moreover, we can only find the optimal strategy parameters with respect to, e.g., historical data, or a hypothetical model. The optimal parameters for live trading are going to be different than what we have picked. We hope that the (small) differences between the optimal parameters and our choices do not severely reduce the profit. So, AlgoQuant checks the stability of the strategy parameters around the chosen values by doing sensitivity analysis to see how the P&L curve changes.
Finally, AlgoQuant uses Monte Carlo simulationto stress test a strategy. We want to ensure (or check) that the strategy is profitable in many plausible situations. For example, the strategy needs to be profitable on average in many possible prices that have similar statistical properties as the ones we use in study before it can go live. Also, to identify the sources of profit, we compute the expected returns of trading a strategy under different (hopefully isolated) risk factors.
AlgoQuant has a collection of trading strategies and models, which can be purchased separately. They are found here: Quantitative Trading Strategies.