ForecastThis is our time series modeling division.
Years of applied machine learning research in collaboration with finance and related industries has led us to push the envelope in robust forecasting and risk optimization. Both in terms of technology and data science process.
Dr. Justin Washtell-Blaise of ForecastThis, at the 2017 Deep Learning in Finance Summit.
Relevant Application Areas
Uncover hidden relationships and drivers.
Anticipate regime changes.
Precisely quantify uncertainty.
Re-balance proactively, not reactively.
Discover novel hedging strategies.
Enforce arbitrary risk preferences and constraints.
Audience & Demand
Discover trends and patterns at all demographic levels.
Optimize pricing and advertising spend.
Stay one step ahead of the market.
Peter Sterling, President of Overland Advisors
Despite continued advances in AI, we advocate caution towards "one-size-fits-all" forecasting technologies. Rather, we produce bespoke solutions which excel in the context they were created for.
We do this by drawing upon proven principles in a variety of disciplines, including: signal processing, information theory, evolutionary algorithms, deep neural networks and ensemble learning... always underpinned by rigorous traditional statistics and data science.
Key Tenets of Our Approach
Don't settle for guesses.
As standard our methods output probability distributions describing every possible combination of outcomes, permitting holistic risk analyses and the quantification of rare or "black swan" events.
Plausibility & Transparency
Work with meaningful models & visualizations - not black boxes filled with obscure math.
In many settings, the patterns underpinning a forecast may be more valuable than the forecasts themselves, leading to deep insights and ultimately better outcomes than relying on machine recommendations alone.
Powerful optimization methods - appropriately marshalled - enable the pursuit of arbitrary objectives, using arbitrary data, under arbitrary constraints.
Stop shaving millimeters off of squared error, and instead discover genuine edge.