Building a deep learning neural network is a complicated and time-consuming task. For some problems it can be easier to re-purpose an existing deep learning model, making minor adjustments to a small number of layers to achieve desired architecture.
In his new book Introduction to Machine Learning, Etienne Bernard gives a nice example of the technique.
We begin with a dataset comprising images of two types of mushroom, with the aim of building a specialized image classifier.
Using a Neural Network for Feature Extraction
We start from the Wolfram ImageIdentify net, which is a 24-layer convolutional neural network trained on around 4000 image types.
net=NetModel[“Wolfram ImageIdentify Net V1”]
In its current form the net is too generalized for our task:
Mushroom
So instead of using the net in raw form, we use the first 22 layers as a feature extractor. This will preprocess each image to obtain features that are semantically richer than simple pixel values. We can then train a classifier on top of these new features, producing a logistic regression model:
This is much better than training directly on the underlying pixel values, which would result in an accuracy of around 44%, worse than random guessing.
Neural Network Transfer Learning
Moving on from Etienne’s feature extraction approach, we next try creating a new neural network by re-using the first 22 layers of the ImageIdentify network, adding two new layers (a LinearLayer and SoftmaxLayer) for mushroom classification:
We re-train the new network using the mushroom training dataset, holding the weights for the first 22 layers at their existing values (i.e. only training the last two, new layers):
Applying Machine Learning in Statistical Arbitrage
In this series of posts I want to focus on applications of machine learning in stat arb and pairs trading, including genetic algorithms, deep neural networks and reinforcement learning.
Pair Selection
Let’s begin with the subject of pairs selection, to set the scene. The way this is typically handled is by looking at historical correlations and cointegration in a large universe of pairs. But there are serious issues with this approach, as described in this post:
Instead I use a metric that I call the correlation signal, which I find to be a more reliable indicator of co-movement in the underlying asset processes. I wont delve into the details here, but you can get the gist from the following:
The search algorithm considers pairs in the S&P 500 membership and ranks them in descending order of correlation information. Pairs with the highest values (typically of the order of 100, or greater) tend to be variants of the same underlying stock, such as GOOG vs GOOGL, which is an indication that the metric “works” (albeit that such pairs offer few opportunities at low frequency). The pair we are considering here has a correlation signal value of around 14, which is also very high indeed.
Trading Strategy Development
We begin by collecting five years of returns series for the two stocks:
The first approach we’ll consider is the unadjusted spread, being the difference in returns between the two series, from which we crate a normalized spread “price”, as follows.
This methodology is frowned upon as the resultant spread is unlikely to be stationary, as you can see for this example in the above chart. But it does have one major advantage in terms of implementation: the same dollar value is invested in both long and short legs of the spread, making it the most efficient approach in terms of margin utilization and capital cost – other approaches entail incurring an imbalance in the dollar value of the two legs.
But back to nonstationarity. The problem is that our spread price series looks like any other asset price process – it trends over long periods and tends to wander arbitrarily far from its starting point. This is NOT the outcome that most statistical arbitrageurs are looking to achieve. On the contrary, what they want to see is a stationary process that will tend to revert to its mean value whenever it moves too far in one direction.
Still, this doesn’t necessarily determine that this approach is without merit. Indeed, it is a very typical trading strategy amongst futures traders for example, who are often looking for just such behavior in their trend-following strategies. Their argument would be that futures spreads (which are often constructed like this) exhibit clearer, longer lasting and more durable trends than in the underlying futures contracts, with lower volatility and market risk, due to the offsetting positions in the two legs. The argument has merit, no doubt. That said, spreads of this kind can nonetheless be extremely volatile.
So how do we trade such a spread? One idea is to add machine learning into the mix and build trading systems that will seek to capitalize on long term trends. We can do that in several ways, one of which is to apply genetic programming techniques to generate potential strategies that we can backtest and evaluate. For more detail on the methodology, see:
I built an entire hedge fund using this approach in the early 2000’s (when machine learning was entirely unknown to the general investing public). These days there are some excellent software applications for generating trading systems and I particularly like Mike Bryant’s Adaptrade Builder, which was used to create the strategies shown below:
Builder has no difficulty finding strategies that produce a smooth equity curve, with decent returns, low drawdowns and acceptable Sharpe Ratios and Profit Factors – at least in backtest! Of course, there is a way to go here in terms of evaluating such strategies and proving their robustness. But it’s an excellent starting point for further R&D.
But let’s move on to consider the “standard model” for pairs trading. The way this works is that we consider a linear model of the form
Y(t) = beta * X(t) + e(t)
Where Y(t) is the returns series for stock 1, X(t) is the returns series in stock 2, e(t) is a stationary random error process and beta (is this model) is a constant that expresses the linear relationship between the two asset processes. The idea is that we can form a spread process that is stationary:
Y(t) – beta * X(t) = e(t)
In this case we estimate beta by linear regression to be 0.93. The residual spread process has a mean very close to zero, and the spread price process remains within a range, which means that we can buy it when it gets too low, or sell it when it becomes too high, in the expectation that it will revert to the mean:
In this approach, “buying the spread” means purchasing shares to the value of, say, $1M in stock 1, and selling beta * $1M of stock 2 (around $930,000). While there is a net dollar imbalance in the dollar value of the two legs, the margin impact tends to be very small indeed, while the overall portfolio is much more stable, as we have seen.
The classical procedure is to buy the spread when the spread return falls 2 standard deviations below zero, and sell the spread when it exceeds 2 standard deviations to the upside. But that leaves a lot of unanswered questions, such as:
After you buy the spread, when should you sell it?
Should you use a profit target?
Where should you set a stop-loss?
Do you increase your position when you get repeated signals to go long (or short)?
Should you use a single, or multiple entry/exit levels?
And so on – there are a lot of strategy components to consider. Once again, we’ll let genetic programming do the heavy lifting for us:
What’s interesting here is that the strategy selected by the Builder application makes use of the Bollinger Band indicator, one of the most common tools used for trading spreads, especially when stationary (although note that it prefers to use the Opening price, rather than the usual close price):
Ok so far, but in fact I cheated! I used the entire data series to estimate the beta coefficient, which is effectively feeding forward-information into our model. In reality, the data comes at us one day at a time and we are required to re-estimate the beta every day.
Let’s approximate the real-life situation by re-estimating beta, one day at a time. I am using an expanding window to do this (i.e. using the entire data series up to each day t), but is also common to use a fixed window size to give a “rolling” estimate of beta in which the latest data plays a more prominent part in the estimation. The process now looks like this:
Here we use OLS to produce a revised estimate of beta on each trading day. So our model now becomes:
Y(t) = beta(t) * X(t) + e(t)
i.e. beta is now time-varying, as can be seen from the chart above.
The synthetic spread price appears to be stationary (we can test this), although perhaps not to the same degree as in the previous example, where we used the entire data series to estimate a single, constant beta. So we might anticipate that out ML algorithm would experience greater difficulty producing attractive trading models. But, not a bit of it – it turns out that we are able to produce systems that are just as high performing as before:
In fact this strategy has higher returns, Sharpe Ratio, Sortino Ratio and lower drawdown than many of the earlier models.
Conclusion
The purpose of this post was to show how we can combine the standard approach to statistical arbitrage, which is based on classical econometric theory, with modern machine learning algorithms, such as genetic programming. This frees us to consider a very much wider range of possible trade entry and exit strategies, beyond the rather simplistic approach adopted when pairs trading was first developed. We can deploy multiple trade entry levels and stop loss levels to manage risk, dynamically size the trade according to current market conditions and give emphasis to alternative performance characteristics such as maximum drawdown, or Sharpe or Sortino ratio, in addition to strategy profitability.
The programatic nature of the strategies developed in the way also make them very amenable to optimization, Monte Carlo simulation and stress testing.
This is but one way of adding machine learning methodologies to the mix. In a series of follow-up posts I will be looking at the role that other machine learning techniques – such as deep learning and reinforcement learning – can play in improving the performance characteristics of the classical statistical arbitrage strategy.
During the 1990’s the advent of Neural Networks unleashed a torrent of research on their applications in financial markets, accompanied by some rather extravagant claims about their predicative abilities. Sadly, much of the research proved to be sub-standard and the results illusionary, following which the topic was largely relegated to the bleachers, at least in the field of financial market research.
With the advent of new machine learning techniques such as Random Forests, Support Vector Machines and Nearest Neighbor Classification, there has been a resurgence of interest in non-linear modeling techniques and a flood of new research, a fair amount of it supportive of their potential for forecasting financial markets. Once again, however, doubts about the quality of some of the research bring the results into question.
Against this background I and my co-researcher Dan Rico set out to address the question of whether these new techniques really do have predicative power, more specifically the ability to forecast market direction. Using some excellent MatLab toolboxes and a new software package, an Excel Addin called 11Ants, that makes large scale testing of multiple models a snap, we examined over 1,000,000 models and model-ensembles, covering just about every available non-linear technique. The data set for our study comprised daily prices for a selection of US equity securities, together with a large selection of technical indicators for which some other researchers have claimed explanatory power.
In-Sample Equity Curve for Best Performing Nonlinear Model
The answer provided by our research was, without exception, in the negative: not one of the models tested showed any significant ability to predict the direction of any of the securities in our data set. Furthermore, our study found that the best-performing models favored raw price data over technical indicator variables, suggesting that the latter have little explanatory power.
As with Neural Networks, the principal difficulty with non-linear techniques appears to be curve-fitting and a failure to generalize: while it is very easy to find models that provide an excellent fit to in-sample data, the forecasting performance out-of-sample is often very poor.
Out-of-Sample Equity Curve for Best Performing Nonlinear Model
Some caveats about our own research apply. First and foremost, it is of course impossible to prove a hypothesis in the negative. Secondly, it is plausible that some markets are less efficient than others: some studies have claimed success in developing predictive models due to the (relative) inefficiency of the F/X and futures markets, for example. Thirdly, the choice of sample period may be criticized: it could be that the models were over-conditioned on a too- lengthy in-sample data set, which in one case ran from 1993 to 2008, with just two years (2009-2010) of out-of-sample data. The choice of sample was deliberate, however: had we omitted the 2008 period from the “learning” data set, it would be very easy to criticize the study for failing to allow the algorithms to learn about the exceptional behavior of the markets during that turbulent year.
Despite these limitations, our research casts doubt on the findings of some less-extensive studies, that may be the result of sample-selection bias. One characteristic of the most credible studies finding evidence in favor of market predictability, such as those by Pesaran and Timmermann, for instance (see paper for citations), is that the models they employ tend to incorporate independent explanatory variables, such as yield spreads, which do appear to have real explanatory power. The finding of our study suggest that, absent such explanatory factors, the ability to predict markets using sophisticated non-linear techniques applied to price data alone may prove to be as illusionary as it was in the 1990’s.
In its proprietary trading, Systematic Strategies primary focus in on equity and volatility strategies, both low and high frequency. In futures, the emphasis is on high frequency trading, although we also run one or two lower frequency strategies that have higher capacity, such as the Futures WealthBuilder. The version of WealthBuilder running on the Collective 2 site has performed very well in 2017, with net returns of 30% and a Sharpe Ratio of 3.4:
In the high frequency space, our focus is on strategies with very high Sharpe Ratios and low drawdowns. We trade a range of futures products, including equity, fixed income, metals and energy markets. Despite the current low levels of market volatility, these strategies have performed well in 2017:
Building high frequency strategies with double-digit Sharpe Ratios requires a synergy of computational capability and modeling know-how. The microstructure of futures markets is, of course, substantially different to that of equity or forex markets and the components of the model that include microstructure effects vary widely from one product to another. There can be substantial variations too in the way that time is handled in the model – whether as discrete or continuous “wall time”, in trade time, or some other measure. But some of the simple technical indicators we use – moving averages, for example – are common to many models across different products and markets. Machine learning plays a role in most of our trading strategies, including high frequency.
Here are some relevant blog posts that you may find interesting:
No doubt many of you sharp-eyed readers will have spotted a spelling error, thinking I intended to refer to one of these:
But, in fact, I really did have in mind something more like this:
We are following an example from the recently published Mathematica Beyond Mathematics by Jose Sanchez Leon, an up-to-date text that describes many of the latest features in Mathematica, illustrated with interesting applications. Sanchez Leon shows how Mathematica’s machine learning capabilities can be applied to the craft of wine-making.
We begin by loading a curated Wolfram dataset comprising measurements of the physical properties and quality of wines:
A Machine Learning Prediction Model for Wine Quality
We’re going to apply Mathematica’s built-in machine learning algorithms to train a predictor of wine quality, using the training dataset. Mathematica determines that the most effective machine learning technique in this case is Random Forest and after a few seconds produces the predictor function:
Mathematica automatically selects what it considers to be the best performing model from several available machine learning algorithms:
Let’s take a look at how well the predictor perform on the test dataset of 1,298 wines:
We can use the predictor function to predict the quality of an unknown wine, based on its physical properties:
Next we create a function to predict the quality of an unknown wine as a function of just two of its characteristics, its pH and alcohol level. The analysis suggests that the quality of our unknown wine could be improved by increasing both its pH and alcohol content:
Applications and Examples
This simple toy example illustrates how straightforward it is to deploy machine learning techniques in Mathematica. Machine Learning and Neural Networks became a major focus for Wolfram Research in version 10, and the software’s capabilities have been significantly enhanced in version 11, with several applications such as text and sentiment analysis that have direct relevance to trading system development:
Continuing a previous post, in which we modeled the relationship in the levels of the VIX Index and the Year 1 and Year 2 CBOE Correlation Indices, we next turn our attention to modeling changes in the VIX index.
In case you missed it, the post can be found here:
We saw previously that the levels of the three indices are all highly correlated, and we were able to successfully account for approximately half the variation in the VIX index using either linear regression models or non-linear machine-learning models that incorporated the two correlation indices. It turns out that the log-returns processes are also highly correlated:
A Linear Model of VIX Returns
We can create a simple linear regression model that relates log-returns in the VIX index to contemporaneous log-returns in the two correlation indices, as follows. The derived model accounts for just under 40% of the variation in VIX index returns, with each correlation index contributing approximately one half of the total VIX return.
Non-Linear Model of VIX Returns
Although the linear model is highly statistically significant, we see clear evidence of lack of fit in the model residuals, which indicates non-linearities present in the relationship. So, ext we use a nearest-neighbor algorithm, a machine learning technique that allows us to model non-linear components of the relationship. The residual plot from the nearest neighbor model clearly shows that it does a better job of capturing these nonlinearities, with lower standard in the model residuals, compared to the linear regression model:
Correlation Copulas
Another approach entails the use of copulas to model the inter-dependency between the volatility and correlation indices. For a fairly detailed exposition on copulas, see the following blog posts:
We begin by taking a smaller sample comprising around three years of daily returns in the indices. This minimizes the impact of any long-term nonstationarity in the processes and enables us to fit marginal distributions relatively easily. First, let’s look at the correlations in our sample data:
We next proceed to fit margin distributions to the VIX and Correlation Index processes. It turns out that the VIX process is well represented by a Logistic distribution, while the two Correlation Index returns processes are better represented by a Student-T density. In all three cases there is little evidence of lack of fit, wither in the body or tails of the estimated probability density functions:
The final step is to fit a copula to model the joint density between the indices. To keep it simple I have chosen to carry out the analysis for the combination of the VIX index with only the first of the correlation indices, although in principle there no reason why a copula could not be estimated for all three indices. The fitted model is a multinormal Gaussian copula with correlation coefficient of 0.69. of course, other copulas are feasible (Clayton, Gumbel, etc), but Gaussian model appears to provide an adequate fit to the empirical copula, with approximate symmetry in the left and right tails.
In a previous blog post I mentioned the VVIX/VIX Ratio, which is measured as the ratio of the CBOE VVIX Index to the VIX Index. The former measures the volatility of the VIX, or the volatility of volatility.
A follow-up article in ZeroHedge shortly afterwards pointed out that the VVIX/VIX ratio had reached record highs, prompting Goldman Sachs analyst Ian Wright to comment that this could signal the ending of the current low-volatility regime:
A linkedIn reader pointed out that individual stock volatility was currently quite high and when selling index volatility one is effectively selling stock correlations, which had now reached historically low levels. I concurred:
What’s driving the low vol regime is the exceptionally low level of cross-sectional correlations. And, as correlations tighten, index vol will rise. Worse, we are likely to see a feedback loop – higher vol leading to higher correlations, further accelerating the rise in index vol. So there is a second order, Gamma effect going on. We see that is the very high levels of the VVIX index, which shot up to 130 last week. The all-time high in the VVIX prior to Aug 2015 was around 120. The intra-day high in Aug 2015 reached 225. I’m guessing it will get back up there at some point, possibly this year.
As there appears to be some interest in the subject I decided to add a further blog post looking a little further into the relationship between volatility and correlation. To gain some additional insight we are going to make use of the CBOE implied correlation indices. The CBOE web site explains:
Using SPX options prices, together with the prices of options on the 50 largest stocks in the S&P 500 Index, the CBOE S&P 500 Implied Correlation Indexes offers insight into the relative cost of SPX options compared to the price of options on individual stocks that comprise the S&P 500.
CBOE calculates and disseminates two indexes tied to two different maturities, usually one year and two years out. The index values are published every 15 seconds throughout the trading day.
Both are measures of the expected average correlation of price returns of S&P 500 Index components, implied through SPX option prices and prices of single-stock options on the 50 largest components of the SPX.
Dispersion Trading
One application is dispersion trading, which the CBOE site does a good job of summarizing:
The CBOE S&P 500 Implied Correlation Indexes may be used to provide trading signals for a strategy known as volatility dispersion (or correlation) trading. For example, a long volatility dispersion trade is characterized by selling at-the-money index option straddles and purchasing at-the-money straddles in options on index components. One interpretation of this strategy is that when implied correlation is high, index option premiums are rich relative to single-stock options. Therefore, it may be profitable to sell the rich index options and buy the relatively inexpensive equity options.
The VIX Index and the Implied Correlation Indices
Again, the CBOE web site is worth quoting:
The CBOE S&P 500 Implied Correlation Indexes measure changes in the relative premium between index options and single-stock options. A single stock’s volatility level is driven by factors that are different from what drives the volatility of an Index (which is a basket of stocks). The implied volatility of a single-stock option simply reflects the market’s expectation of the future volatility of that stock’s price returns. Similarly, the implied volatility of an index option reflects the market’s expectation of the future volatility of that index’s price returns. However, index volatility is driven by a combination of two factors: the individual volatilities of index components and the correlation of index component price returns.
Let’s dig into this analytically. We first download and plot the daily for the VIX and Correlation Indices from the CBOE web site, from which it is evident that all three series are highly correlated:
An inspection reveals significant correlations between the VIX index and the two implied correlation indices, which are themselves highly correlated. The S&P 500 Index is, of course, negatively correlated with all three indices:
Modeling Volatility-Correlation
The response surface that describes the relationship between the VIX index and the two implied correlation indices is locally very irregular, but the slope of the surface is generally positive, as we would expect, since the level of VIX correlates positively with that of the two correlation indices.
The most straightforward approach is to use a simple linear regression specification to model the VIX level as a function of the two correlation indices. We create a VIX Model Surface object using this specification with the Mathematica Predict function:The linear model does quite a good job of capturing the positive gradient of the response surface, and in fact has a considerable amount of explanatory power, accounting for a little under half the variance in the level of the VIX index:
However, there are limitations. To begin with, the assumption of independence between the explanatory variables, the correlation indices, clearly does not hold. In cases such as this, where explanatory variables are multicolinear, we are unable to draw inferences about the explanatory power of individual regressors, even though the model as a whole may be highly statistically significant, as here.
Secondly, a linear regression model is not going to capture non-linearities in the volatility-correlation relationship that are evident in the surface plot. This is confirmed by a comparison plot, which shows that the regression model underestimates the VIX level for both low and high values of the index:
We can achieve a better outcome using a machine learning algorithm such as nearest neighbor, which is able to account for non-linearities in the response surface:
The comparison plot shows a much closer correspondence between actual and predicted values of the VIX index, even though there is evidence of some remaining heteroscedasticity in the model residuals:
Conclusion
A useful way to think about index volatility is as a two dimensional process, with time-series volatility measured on one dimension and dispersion (cross-sectional volatility, the inverse of correlation) measured on the second. The two factors are correlated and, as we have shown here, interact in a complicated, non-linear way.
The low levels of index volatility we have seen in recent months result, not from low levels of volatility in component stocks, but in the historically low levels of correlation (high levels of dispersion) in the underlying stock returns processes. As correlations begin to revert to historical averages, the impact will be felt in an upsurge in index volatility, compounded by the non-linear interaction between the two factors.
The Volatility ETF Trader product is an algorithmic strategy that trades several VIX ETFs using statistical and machine learning algorithms.
We offer a version of the strategy on the Collective 2 site (see here for details) that the user can subscribe to for a very modest fee of only $149 per month.
The risk-adjusted performance of the Collective 2 version of the strategy is unlikely to prove as good as the product we offer in our Systematic Strategies Fund, which trades a much wider range of algorithmic strategies. There are other important differences too: the Fund’s Systematic Volatility Strategy makes no use of leverage and only trades intra-day, exiting all positions by market close. So it has a more conservative risk profile, suitable for longer term investment.
The Volatility ETF Trader on Collective 2, on the other hand, is a highly leveraged, tactical strategy that trades positions overnight and holds them for periods of several days . As a consequence, the Collective 2 strategy is far more risky and is likely to experience significant drawdowns. Those caveats aside, the strategy returns have been outstanding: +48.9% for 2017 YTD and a total of +107.8% from inception in July 2016.
You can find full details of the strategy, including a listing of all of the trades, on the Collective 2 site.
Subscribers can sign up for a free, seven day trial and thereafter they can choose to trade the strategy automatically in their own brokerage account.
The Futures WealthBuilder product is an algorithmic CTA strategy that trades several highly liquid futures contracts using machine learning algorithms. More details about the strategy are given in this blog post.
We offer a version of the strategy on the Collective 2 site (see here for details) that the user can subscribe to for a very modest fee of only $149 per month. The Collective 2 version of the strategy is unlikely to perform as well as the product we offer in our Systematic Strategies Fund, which trades a much wider range of futures products. But the strategy is off to an excellent start, making +4.4% in June and is now up 6.7% since inception in May. In June the strategy made profitable trades in US Bonds, Euro F/X and VIX futures, and the last seven trades in a row have been winners.
You can find full details of the strategy, including a listing of all of the trades, on the Collective 2 site.
Subscribers can sign up for a free, seven day trial and thereafter they can choose to trade the strategy automatically in their own brokerage account, using the Collective 2 api.
