I have received the following questions from a reader and I’m happy to publicly reply, considering that the answers may be of general interest. Thank you for asking.
1. How exactly does the daily and weekly forecast work? can you give an example?
The daily, the weekly and the monthly forecast are based on three different models (one each) that, following the Artificial Intelligence jargon, are nothing else than very large matrices or tables, not very different, in structure, from an excel sheet. The columns are composed of financial and economic instruments as stock market indices, rates, forex, commodities, economic indicators and so on.
The sheet is then “trained”, meaning that each row of data is associated to a desired output: it’s a sort of diagnosis, where the optimal output is defined. Finally, the model is fed into a backpropagation neural network, which generate all the correlations necessary to provide a result from never seen before data. The whole process is not very different from a medical diagnostic tools, that correlates the values of many exams to a possible disease or to meteorological forecast: the data (i.e. today’s values of the inputs) are evaluated by the pattern recognition engine (the neural network) and the output is generated.
The three models (daily, weekly and monthly) are different: the daily is composed almost completely from financial data, the monthly has a large presence of econometric data and the weekly is a mix of the two categories. Also the length of the three databases are very different, going back into the past for as much as I have been able to collect the raw data.
2. Do you provide exact entry or where the spx will close for the day or next day?
The models generate five indicators:
– Forecast Bars, that is a projection into the future of the value of the SPX. This is a real forecast, as the model outputs the predicted values of high/low/close for the next 24 bars (around one month for the daily model, about six months for the weekly and two full years for the monthly one). This prediction is very volatile, meaning that it may change from a day to the next quite consistently, and it is sometimes very precise, other times totally wrong. What’s more interesting in this indicator is the evaluation that the model does of the implicit cyclicity that the data are expressing and the overall behavior that the index is going to have in the near future.
– Target, that is an evaluation of the value of the index at the end of the present move. This also is a real forecast, less volatile than the bars.
– Stop is a not a forecast, but a reading of the instant situation and an evaluation of the value that the index should reach to reverse its current position.
– then we have Signal and Position. First came Signal, a simple mark of the extremes in the index jigsaw. Then it was evolved into the Position, a more complex indicator that not only marks the extremes, but follows the move of the index step by step. Finally, just few weeks ago, the Signal has been redesigned to mark extremes with a hierarchy of major and minor tops and bottoms.
3. What percentage accuracy is the forecast? And for how many years?
The whole system is under development since the spring of 2013, so about four years ago. It took a couple of years to start walk on his own legs and the spxbot.com web site was launched in November 2015. A crucial improvement was introduced in the fall of 2016, around six months ago. It is under continuous evaluation and development. The model sometimes reads the market with astonishing precision, sometimes goes dumb. As it has been highlighted by some users, it is very reactive (it has been designed to be adaptive and responsive) and adapts to changing conditions very fast.
Chart of the daily forecast one day ahead: blue the SPX index, pink the forecasted value.
A final word: originally this research was conceived to support my own investment activity, not strictly for trading. My horizon has always been medium to long term, something that is classically defined as “position trading”, or trying to profit safely from major market moves.
The comments are open, so that if any current subscriber want to contribute with her/his experience, I’d be happy to hear from you. Your opinions may be invaluably precious to push development toward new goals.