Mind Foundry Horizon overcomes important challenges across the sequence of procedures involved in time-series forecasting.
The first of these challenges is intelligently matching a task-goal with a pre-integrated workflow.
Drawing on years of Mind Foundry experience, Horizon provides a library of curated workflows, each one suited to a challenge that quantitative analysts regularly confront.
A distinguishing feature of Horizon’s technical design is the composition of a forecast based on a set of separate machine learning models, one for each future point in time plotted in the forecast.
This contrasts strategies of classical forecasting, but it makes enormous sense when you consider how the relative influence of features changes over time. Overfitting is a notorious source of trouble in financial time-series forecasting. Horizon’s approach to signal selection tackles that problem with a new paradigm.
As workplaces become more collaborative over time, machine learning that adopts a “black box” means to its ends runs into a problem: Trust. It’s not just that Mind Foundry has designed Horizon to be explainable.
Mind Foundry has designed Horizon to nearly explain itself - with intuitive displays and time-saving, graphically rich automated reporting.
In developing Horizon, Mind Foundry consulted with a set of quantitative analysts and strategists (“Quants”) in the financial sector. Responding to their requirements, Mind Foundry built the platform with use cases centred on the exploration and scenario-development aspects of their work. Workflows, and the statistical machine learning principles underpinning them, anticipate central challenges of analyzing financial time-series data: periodicity, stationarization, overfitting and wide variations in the size of data sets.
In those areas where you can add value through your expertise, intuition and creativity, Horizon offers significant configurability. In those areas where you would like to save some time (such as backtesting and reporting) Horizon radically compresses the time required.
Want to explore small data sets? Horizon helps you generate forecasts accompanied by a principled assessment of model uncertainty.
Want to wrangle large data sets? Horizon is optimized for daily data, but adaptable to a wide range of data sampling rates and sizes.
Want to deploy on-premise or in the cloud? Have it your way. See something missing that would make Horizon even more awesome for you? We’re listening.
To arrange a discussion or a demo, please get in touch