A Quantitative Study for Volume in Futures Trading

Last modified date

How to Understand the Role of Volume in Futures Trading Systems

If there’s a piece of disinformation about trading, that may come, mostly, from honest traders, but it is wrong, is probably all the misconceptions about Volume and its role for creating trading systems.

Let me go through this, so you can see what I am talking about, recently, while watching videos on Youtube I’ve stumbled upon an ad from a self-proclaimed Futures Trader, he said, and I’m quoting this literally: If you trade without Volume you are going to end up bankrupt since you can’t see what professionals are doing.

By the way, this was an Spanish Ad, don’t bother looking for it if you don’t speak Spanish.

The thing is, that it made me think, I only use Volume for a couple systems, one using Time & Sales and a couple using On Balance Volume. So I already knew that the claim was just a marketing stunt in order to sell some shady material, but anyway, if people believe that sort of information, maybe I need to make and share some data mining studies about this.

How Professionals Trade

There’s another huge misconceptions about trading firms, hedge funds, CTAs and companies alike.

People usually tend to think that they only use Volume for making trades, and they can’t be more wrong, first of all, not everyone trades the same way, and I’m not talking about the average retail trader who was a narrow set of options when it comes to create trading strategies.

Let’s set aside the fact that some companies and funds do not seek for a single price change to make a profit from the market, people who make Stat-Arb are the best example, let’s talk a little bit about information flow and how trading models are created.

Some people indeed use volume for making trading decisions, that’s correct, but also a wide variety of techniques too, from Data Mining, Dual Momentum Models, Deep Learning to Seasonal Trading.

Some Hedge Funds and CTAs use alternative data, like the video above, to make predictions about the price of commodities using satellite imagery of crops.

Trading Systems in the Futures Market comes in many shapes and colors, according to how they want to approach the market.

In the current days most Funds are looking forward to create more Quantitative Models implementing Machine Learning Techniques, and for that, they use huge datasets in most cases.

In summary for this part, you can’t use a single factor in most cases, maybe for long-term models, due to this fact (we are going to learn why a bit later in this post): The higher the Time-frame, the higher the correlation between a single factor and the next price movement will be.

This is due to market’s noise, and we can learn how to measure it creating models.

Measuring Volume relevance in today’s market

We have the following hypothesis, Volume is everything in the market, so it must be the most predictive factor for the next price change according to that statement.

Well, let’s find out.

For this study I’m going to run a few simple tests based on data mining techniques, in this case, I want to measure the correlation coefficients between some values and features against the next price. 

This means, we have a list of given time-series, such as some indicators values, volume ratios and price ratios, and we are going to measure the correlation of these series against the next price, or said in other way, these values are going to have a lag of 1 period below.

This way, we want to measure the relationship between changes in volume and other factors against the price movement.

The highest correlation will be the ones that have more value as features for modeling strategies.

In order to do this, we have many options, but in this case I don’t want to do anything complicated, since it doesn’t need to be that way to have effective.

I’ve exported a CSV file containing some values from indicators based on price and volume, volume ratios and price ratios along with the next return labeled as “1” and “-1” and run some correlation tests for each row.

Keep in mind that this contains every bar on the market and its next return, since someone already asked me that.

You can perform this kind of tests using a variety of software and programming languages, everything from C#, Weka or Python. I’ve done this using Java, since I had similar codes already done for different models that I trade.

If you are interested in learning about Data Mining check this website, it has a lot of resources for free.

Study Results

In this study, or analysis, I have selected five Futures Contracts along with some different time-frames, having a decent sample from @ES (S&P500 Futures) to @S (Soybean Futures), with time-frames from 100 Tick Charts to Daily Bars.

This way, we can analyse this from a wide point of view thanks to different contracts and time-frames, with a green background, volume based features:

Volume and other features in the futures market
Most Correlated Features in the Futures Market – Volume Features are marked on green.

I need to make a note here before continuing, or better said, a couple notes.

First, I didn’t focused the study on the Indicators part, this means, there are indicators with a higher negative correlation than the ones displayed with a positive one. I only have taken notes in case that we find a negative volume correlation.

And second and most important, Vol(30) and Vol.Std Dev(30) are Volatility measures, not Volume ones.

Being that said, we can extract some useful information here, first of all.

Volume is only a relevant feature in small Time-Frames, no point in use Volume only if you are using a Hourly (or 60 Minute, if you are a Tradestation user like me) chart, neither a Daily one.

Second, its relevance also depends on the contract, as you can see, it is more relevant in Gold than in the S&P500 Index.

And last, I didn’t put up the correlation values since they are low, that happens due to market’s noise, the smaller the time-frame, the higher the noise is. This explains why we need to add more rules when creating a trading system using a 10 minute chart than when we create one on a Daily Chart. Makes sense, right?

I think this states a great point also on Time-Frames where Volume is not that relevant, people may tend to think that Volume is the only factor, but there is one factor that is usually being forgotten by almost every trader, volatility.

Some Stats: For the 100 Tick Chart and the 15 Minute one Volume counts as 12 out of 25 Features. In the daily chart volume counts for 2 of 25 Features.

If we focus on creating winning trading systems, we need to model volatility along with price and indicators, we can use volume too.

As a small bonus, have you noticed how price appears to be relevant in some scenarios?

In this case, Japanese candlesticks patterns are still a thing, and I don’t believe in them, but that doesn’t mean that we can quantify some patterns based on simple ratios like High[1]/High[3], and they sure work better than the old stuff, markets tend to evolve.

As always, hope you enjoyed this article and it helps you to become a better trader.

Víctor – Follow the Edge.