Wallaroo | Trading Analytics Solutions
  • White Twitter Icon
  • White LinkedIn Icon

© 2019 by Wallaroo Labs, Inc.
Made with  in NYC.

Trading Analysis

Trading strategies rely upon continuous and predictive analytics of massive amounts of real-time trading data.

We can help.

Sample Use Case

Trading Scenario Analysis

Opportunity

Real-time predictive applications like algorithmic trading require continuous and agile training, testing and deployment of automated machine models (trading strategies).

This is done by replaying logs of historical trade events and examining how many new potential strategies would perform in the context of the historical data relative to a desired business outcome (e.g. expected return),  i.e. "backtesting" the strategies.

The best strategy is then be selected for production. In some organizations this process is repeated daily.

Challenge

There are three main technical challenges to overcome:

 

  1. Speed. The backtesting needs to run thousands of times faster than the real-time rate of data that is being replayed. For example, testing a trading model against the previous years worth of trading data should happen within a few hours. Similarly, testing 1000s of potential new strategies against the previous day's data requires overnight processing before trading starts the next day.
     

  2. Cost. The backtesting needs to run cheaply but with guarantees. Because each trading strategy is accumulating state as new trade events are played into it, a failure can result in errors or not having the best information to select the most fit strategy. In addition, the cost of running the analysis on large numbers of fixed servers can reduce the business benefit of continuous re-training because it eats into any increased revenue.
     

  3. Scalability. The system has to support multiple simultaneous data scientists (quants) testing ideas simultaneously, without constraining capacity.

How Wallaroo Helps

Wallaroo provides easy solutions to each of these challenges.

 

Wallaroo allows for on-demand deployment at the necessary scale. So for example, depending on the size of the data, the number of trading models to be tested, and the required time-to-completion, the appropriate size cluster would be created on the fly and the processing would be done on-demand.

 

This could happen on a per-quant basis in the cloud so there is no constraint on capacity.