Time Series Terminal Whitepaper
Brief Solutions Ltd, May 2022
The Problem
Is time series A predictive of time series B?
i.e. given all information about A's value up to time , is it predictive of the conditional distribution of B's value at a future time , with being the prediction horizon?
Is this relationship causal?
i.e. the presence of A makes the prediction of the future of B better, its absence makes it worse.
The Solution
Given a pre-configured collection of time series data, we designed a computational engine to compute the predictive power of any series A for any series B, for a given prediction horizon. When we collate the results we obtain a causal graph that represents the strengths of predictive power between them.
For example, following the close of Mon 23 May 2022, the computational engine updates all causal relations (for 2 day ahead prediction). The vicinity around the Invesco Nasdaq-100 ETF would read like
The engine employs a group of trained models to make prediction for any target time series. Each model eventually produces a single output: "up", "down" or "undecided" for the direction of movement for a given horizon. When grouped together, with the Invesco Nasdaq-100 ETF as the forecast target,
-
For Tue 24th, 20% predict up, 70% predict down, 10% undecided;
-
For Wed 25th, 40% predict up, 60% predict down, 0% undecided.
The net predictions, in terms of the percentage of models predicting up minus that predicting down are thus
-
For Tue 24th, 50% predict down;
-
For Wed 25th, 20% predict down.
Backtest
As one exercise, if we use the above net prediction as size for hypothetical trading, under ideal condition of no trading cost, the performances would be:
For Invesco Nasdaq-100 ETF,
For SPDR S&P 500 ETF,
For SPDR Dow Jones Industrial Average ETF,
For US 10-Year Interest Rate,
