Posted 2 years ago | @cellis
I started building Data Investors as a web-based application for my personal use in investing over 3 years ago. After friends and family requested access, I opened the site's content to create a community around data-driven investing. Once registered, you'll find a trove of useful data to guide your investment decisions, including data about individual companies, price history, macroeconomic data, news, social signals, sec filings, analyst recommendations and price targets, and much, much more. I'm confident you won't find a site with richer, more accurate, or more timely data.
The amount of data available, however, can leave even the most seasoned data professional in a state of "Analysis Paralysis" - where the sheer volume of possible analyses and variables results in an infinite research loop. This is where the predictive modeling aspect of Data Investors comes in. I've built a backend that continuously retrains and optimizes predictive models to achieve the highest predictive accuracy possible (evaluated using a multi-year backtesting algorithm I developed - more to come in a later blog post). This iterative process tests which data (and how much, among other parameters) is most predictive of stock price movement and then selects the best performing model for use in daily (production) predictions. Each model has its own detail page where you can see exactly how the model actually performed historically (or would have performed for newly released backtested models), as well as what data the model relied on (and the relative importance of each variable).
Finally, based on the data and model predictions available, you can create a "Pick" representing your sentiment on any stock. A pick is a point-in-time prediction of a stock's future price change, consisting of 1) a Ticker, 2) a price change (including direction, condition and % change), and 3) a max holding period. For example, you may believe:
a) Apple (AAPL) will increase by at least 10% in the next 10 trading days
b) Tesla (TSLA) will decrease by at most 5% in the next 5 trading days
c) Overstock.com (OSTK) will increase by at most 15% in the next 242 trading days
d) Intel (INTC) will decrease by at least 2% in the next 5 trading days
Each of these would be a different Pick that, once created, would be tracked and scored by the site's backend based on the time the Pick was submitted (a future version of the site will even present you with different trades you could make based on your belief, incorporating derivative trading strategies). The scoring process simulates real-world investing by creating a simulated trade based on your pick and then exiting that trade when either a) the price change condition is met, or b) the holding period expires, whichever is sooner. The picks are then scored based on the annualized return of each pick, with comparisons to total-market performance available for reference (after all, the real test is how your picks perform relative to the alternative: not choosing stocks and just buying the index).
As an example, if Pick (a) above were submitted on a Sunday evening, the backend would create a simulated trade purchasing AAPL at the open the next day (Monday), and then would track that trade thereafter, selling either a) when the stock's price reaches +10% relative to the purchase price, or b) at the close on the 10th trading day (Friday of the following week, assuming no weekday market closures). The annulized return would then be calculated based on the realized return / holding period. If the realized result was a sale for a 10% profit on the last holding day, the annualized return would be 1.1^24.2 (where 24.2 equals the number of times this strategy could be repeated in a given year: 242 trading days / 10-day realized holding period) = 1003%. If every stock in the market increased by 15% over that same time period, then a reference return of 2943% would be shown, suggesting that, despite the great gains, you would have been better off just buying the index.
Over time, as the actual results of your Picks are realized, you earn an investing score based on your performance. Additionally, other users can follow you to see your picks, knowing that your investing score is based on real selections you made in real-time (as opposed to a subset of your picks that you cherry-picked, or fake picks created after-the-fact). Unlike other investment-discussion sites, you can look at a user's track record before giving weight to their comments or picks.
As this was originally created for personal use, before any other users existed, I use the model predictions in conjunction with the scraped data to guide my investment decisions. I'm currently still iterating on the best way to use the models and data in my personal trading. At the highest level, I use this site for any research I want to perform on a stock, knowing that my site aggregates data from any and all sources I'd be interested in (if a new source ever arises, such as the proliferation of WallStreetBets, I simply add it to my scrapers).
One way I've tested using the site is to choose a single model and purchase that model's top 10 predictions each day. Backtesting this strategy demonstrated a realized return of 8% per month over a 3-year investing period between 2019 and 2022. A downside of this strategy, however, is that it picks a single model, resulting in a one-direction investment strategy that performs well when the market trends in the same direction and performs poorly otherwise.
As a result, the way I've used the site is to start by looking at model predictions for the entire market to first determine the trade direction (long vs. short vs. directionally neutral). I then sort the industry / sector by average prediction and perform my own analysis on the 3 stocks with the highest predictions. Finally, I calculate my expected return based on the model's historical performance, and use this in conjunction with my analysis to make final investment decisions. Thereafter, I set up an email notification to alert me if the predictions for any of the selections I made change substantially.
@cellis | Dec. 15, 2022, 10:34 p.m.
There are no comments yet.