Original article was published on Deep Learning on Medium
Studying volatility, uncertainty and news—Case study #1
In this post, we’ll share some of our thoughts on how to study market volatility and news. We’ll give you some pointers on how we at Predic.to deal with volatility and uncertainty in order to understand risk involved at any given time. This is a continuation of our series of posts about Investing + Deep Learning.
Volatility is a very important factor of the markets, and it’s wise to always be aware of it. Volatility of specific sectors and specific stocks can give you clues on what to expect next. News and events happening around the world play the most important role of course. Investors react fast to important events. Sometimes they overreact, sometimes underreact and most of the times they don’t react at all. The hardest part is understanding how different news events can affect various company sectors and when this occurs.
Let’s do a case study for a recent period of 2 weeks, May 23rd — June 6th. In the table below, we can see that lots of stocks from different sectors have been very volatile during that period. A first observation is that airline stocks moved significantly. A second observation is that Slack stock even though it had significant moves (standard deviation ~7%), price remained at the same levels (+0.46% move).
Similar information can be seen in sectors volatility table for the same 2 weeks period. To track sectors, we use several ETFs. E.g. JETS tracks the airlines industry and we can see significant increase there.
After identifying volatile stocks/sectors, it’s always a good idea to understand why those stocks have been moving so much.
For example, let’s start with Slack and see what happened. First of all, Slack expected to release its quarterly earnings report on June 4th after market close. Stocks are usually volatile before and after that period. The same thing happened here. The days leading to earnings, lots of news about what to expect came out, some analysts said some good things expecting good results and Slack in general seemed to benefit from remote working due to coronavirus. Everything looked good.
Then suddenly on the day of earnings (June 4th), Slack stock lost -5% during the market session just before earnings announcements, and another -15% until next day’s opening. Even though they announced a deal with Amazon to use their product and earnings results looked ok (according to the headlines). It seems worries about competition with other companies played a part on this.
The main point is that the stock ended up being at the same level where it was 2 weeks ago and all this volatility was just noise from expectations.
In this case, our forecasting model did a good job staying calm in its prediction but didn’t predict the volatility.
Now let’s dig deeper in American Airlines (AAL) stock movement.
We can see obvious ups and downs and serious volatility leading to ~60% rise in the last few days. Especially on June 4th there was a +41.1% move in a single day! Here are some insights on what happened on that day’s session:
After looking around in the trending news (table above), it is clear that news about American Airlines reports about slow and steady rise in domestic demand did the trick. Airlines sector got a boost, but especially AAL, if we compare against JETS ETF.
Now, once we have a good picture about recent volatility and why it happened it would be nice to have a way to track expected future volatility or even better calculate an estimate of volatility and future stock direction.
A simple way to do this at the market level is by keeping an eye on VIX index which measures expected market volatility of the next 30 days as this is implied by options data — which means how investors expect the market to fluctuate (more info here).
There are more ways though. To do this, we experimentally use deep learning models based on our data.
For the AAL case, our model gives us a forecast that we can use to calculate expected volatility in the next 2 weeks and some price movement. Also it gives us a level of uncertainty about this forecast. In this case it gives us 5.8% predicted volatility with +/-16.8% forecast uncertainty which is pretty high.
10 days ago our forecast looked like this
The model didn’t do a good job at predicting the move but it was pretty uncertain about its prediction at +/-30% uncertainty. Prediction stayed within uncertainty bounds for most of the time, until that last huge +40% jump which is indeed outside the 95% percentile of this stocks movements, something you don’t see every day.
To compare volatility levels, below you can see our forecast for Apple (AAPL) at that day with low uncertainty at +/-5.4% and pretty good forecasting accuracy. The volatility of the predicted forecast was 0.7% and the actual volatility during that period was 1.1%.
The difference is that AAPL is a much more stable stock than AAL recently.
As a last example, let’s go through the list of the least volatile stocks during the same period and have a look at our forecast from 2 weeks ago for one more stock.
Walmart (WMT) seems to be the least volatile for that period. Indeed there were no big moves.
How did our forecast, uncertainty and predicted volatility look 2 weeks ago for WMT?
Predicted volatility was at 0.6% ( actual was 0.7%), forecast uncertainty was pretty low at +/-2.4% and forecast was very close to actual movement. In this case, our model was able to identify the low volatility nature of WMT.
There were no big news for Walmart during that period, or it’s actually much better to say that there were no news that could move the price significantly. The only news that seemed important were on June 3rd some headlines about mobile payments in India related to Walmart’s PhonePe and stores closures news related to protests on June 1st. Nothing moved the price too much though.
And we can keep going with our research for as long as we like. There are hundreds of moves happening daily waiting to be explained.
This was just one case study that started from a recent move that caught our eye. We started from Slack news, then moved to airlines sector, then we navigated across industries to end up checking forecasts for Apple stock and Walmart’s moves in India! Wow!
In this post, we wanted to give you a very brief idea of how we study news and a sneak peek on how we experiment with the concepts of volatility and uncertainty in our service.
There is a long way to go and we have just scratched the surface. We’ll come back soon with more case studies and insights.