S2O AI Buy/Sell signals
Many investors are using technical analysis to trade stocks market instruments. Most of the technical indicators are widely used and become a self-prophesy – 200 days moving average would be a good example. It makes investors constantly search for new indicators or combinations of indicators to find unique opportunities and perfect market timing for entries and exits.
The more indicators investor uses, the more skills and time required from a human investor to find a right instrument and to trade it profitably. At the same time, the task can be described as “pattern recognition” which makes it a good fit for machine learning algorithms.
Several different machine learning models have been intensively tested with different sets of historical market data. As of summer 2018, the best models recognize price changes by generating Buy / Sell signals with more than 85% accuracy.
S2O AI signals are available on stock2own charts
You can see AI signals on price charts. They presented as letters B or S, representing Buy or Sell signal and available on both - static and dynamic - charts.
Note that as of this writing we can compute AI signals only for US securities.
All AI signals generated by stock2own are currently based on technical indicators. We have experimented with many different models and as of today, the most promising models have more than 300 data points and include information about a particular stock and market in general (S&P 500 and other indexes, treasury yield rate and more). We trained AI models with 17+ years of historical data for more than 3000 US stocks. We selected most recent history, starting with year 2000 and, as you know, it covers couple of bear markets (year 2000 dot-com bubble, 2008 financial crisis) as well as longest so far bull market in history.
We compute AI signals daily, after market closed. It allows you to incorporate this knowledge into your daily routine and use freshly baked alerts before market is open.
AI signals can be used for
AI signals can be used to trade stocks and stocks derivatives. For instance, if you are expecting stock to rise, you may want to buy a stock, buy a call option on the same stock or sell a put option on the stock. All these types of trades require underlying security analysis and S2O AI signals can really help you identify great opportunities.
Machine learning models signals could be mixed with different strategies to produce better results. For instance, models “mistakes” could be partially neutralized by using stop-loss orders to auto close positions and conditional orders to minimize mistakenly open positions.
Buy low, sell high
"Buy low, sell high" – this is the main objective all S2O AI models were trained with. It might look quite contrarian on the chart though, because you will see that stock is going down and all of a sudden AI models generate a Buy signal. Think about it as an early indication of a potential pivot point, but make no mistake – no one can predict the future with a 100% probability. Even an Artificial Intelligence. Especially in a stock market where price is often driven by emotions and not by logic or common sense.
In order to make AI signals more practical, we are using conditional orders with stop losses.
Conditional order allows you to execute transaction only when certain conditions are met, which is extremely useful to deal with uncertainty. Let's say that we have a Buy signal. Buy signal means that the stock price is expected to rise and therefore we want to place a Buy long order. Before doing that we need to make sure of two things:
- We want a confirmation that stock price is really going up.
- We want to mitigate losses in case if everything goes wrong.
To solve both problems we are using ATR – Average True Range indicator. Essentially ATR shows the daily price change for the last several days (usually for the last 14 days), it provides a good estimate of how much the price usually changes during one trading day. For example, if the stock price is $100 and ATR is $2, we know that it usually swings up or down no more than $2/day, in other words the price fluctuation during the day most likely will be between $98 and $102.
To test AI signals we are using 10% of ATR to confirm the price move and 10% of ATR to mitigate the risk. So, in our example, where current stock price is a $100 with a daily range of $98 - $102, ATR is $2 and we expect price to move up (got Buy signal), we would place an order for automatical execution the next business day saying that we want to buy a stock if the price make higher high - move to $102.20 or higher (confirmation of the move up) but not higher than $102.40. That second piece is supposed to protect us from possible "gap moves", where price can jump overnight and we may end up paying too much for a stock.
To mitigate losses we also want to place a stop-loss order which will close our position (sell the stock) if the price drops below $97.80 (low price of the day minus 10% of ATR). This second order must be placed only if the first part of the order went through. In other words, we do not want to sell a stock if we did not buy it.
Most online brokerages allow you to place such orders via OCO Bracket orders. OCO states for "One Cancels the Other". Please, read your brokerage documentation for more details.
S2O AI signals testing
- To better understand results, a single stock is traded in the virtual portfolio in each test run.
- Only one open position allowed. It means that if portfolio has an open position and another "Buy" signal generated by a model, no changes will be made in the portfolio.
- The best results we got when both short and long trades are allowed. It means that sell signal may either close an open position or initiate a short trade.
- All simulations performed on stocks that were NOT used for models training.
- For illustration purposes, portfolio simulator has $10,000 initial balance. This value is irrelevant but makes test results somewhat similar to real life and easier to understand and compare.
For illustration purposes we selected a stock that had a bit of a downtrend and an uptrend. Here are the results for the US:F (Ford Motor Company):
As you can see, model was robust enough to generate reasonably well signals in any trend, which resulted in a nice 55% portfolio value growth from 2013 to 2018 even for the stock which price gained literally 0 for the same period of time.