I can model a quantitative trend and forecast it into the future. This assists with predicting the price of an investment instrument (to assist with buy or sell decisions); or to predict the future state of a system, such as an economy, the operations of a business or a natural setting. I can use a few tools to achieve this, although I am fairly inexperienced at doing this professionally:
ARIMAX: Autoregressive Integrated Moving Average modelling, possibly with exogenous cofactors
Luno Bitcoin candlestick close prices
I have an advanced method in mind:
Aggregate the trend to a less-frequent period (e.g. monthly). Forecast well into the future (e.g. a few months).
Interpolate the aggregated trend to the more-frequent period, and take the difference between the noisy data and the interpolated trend. Apply time-series modelling on those errors, forecast, and add on top of the long-term trend.
A third-stage regression can be used, for values just ahead of the present time.
This method above captures long-term trends as well as short-term trends. I can implement this analysis for a time-series of your choosing (for example, a fund on investing.com, or from data that you provide), and provide two or three graphs of the forecast, with a steady-state smooth trendline. This offering would take about 8–12 hours to set up, and then we could automate it on a daily schedule, for about 3–4 hours per month.
GARCH: Generalized Autoregressive Conditional Heteroskedasticity model
For example, sales of a product may drop off as it is discontinued — the standard error in the trend would be smaller for lower levels of sales, and larger for higher levels of sales. The GARCH model can validate the last observed level of sales reported, by looking at the forecast error on the last datum. And the model can forecast into following periods.
VARIMA: Vector Autoregressive Integrated Moving Average
This method integrates multiple variables, to interact in a system of time-series equations. The model lets us predict multiple variables simultaneously. The root mean squared error is minimized across the equations I think. This method is necessary when there are multiple variables, for example, an economy, or the influence on stock prices. However, I think it only works when the period (frequency) for each variable is the same. I am interested in applying this method in a project.
I can use a social accounting matrix (SAM) to study the effect of an injection in an economy, across various industries. I only have about 25 hours of work experience with this so far. I am keen on learning about CGE modelling.
I have tutored Dynamic Optimization to Masters and PhD students at the University of Cape Town. I own a copy of Chiang's Elements of Dynamic Optimization.
This model can be used to find the shadow price of a natural resource over time.
I can consult on a problem, draw up equations, or help to create modelling software for you.
I can partner with you in developing an algorithmic cryptocurrency day-trading bot. I have a strategy based on past prices and the Fear and Greed index, but I have not spent time on other exogenous variables or back-testing. At the moment, my lack of capital doesn't really overcome the running cost, including labour. This needs to be a long-term partnership.