Monday, June 19, 2017

Predictions on the Stock Market

A simple prediction about the US stock market - I am anticipating a crash in US equities especially those in the tech sector.  There are several underlying causes of a large concentration of investments in what are referred to as the FANG stocks (Facebook, Amazon, Netflix, Google).. in fact, the sector is responsible for a whopping 40% of the S&P's gains year to date.

As a consequence of this concentration, tech stocks, and this includes the semi-conductor sector, are highly overvalued, in some cases 20 or 30 times, on par with the crash of the nineties with the dot com bubble.  Meanwhile, other sectors have fallen, including energy, into far oversold territories (based on factors such as the Relative Strength Index, Earnings Per Share, and capital return on investment caluclations).

So why is this happening?

Among many factors, the one which I believe will cause a crash is the transition from institutional investing from humans to machines. 

Increasingly, machine learning is used on the stock market, to trade at extreme high frequency, and oftentimes on private, secret exchanges known as Dark Pools.

So how do these artificially intelligent traders work?  AI traders are concerned primarily with price movement, which is a lot of ways, is in the language of the machine - calculus.

Firstly, let's define the price of a stock.  A stock's price is the sum of all investor sentiment in the value of a business. 

All that complexity, the factors the influence that sentiment and factors influencing those factors, creates a feedback hierarchy of immense complexity, where interrelated variables can have nonlinear influence on each other, making for a seemingly unpredictable movement in price, or what is commonly called the "random walk" hypothesis of price movement.

The opposite is true.  There are factors that influence the price, like the profitability of the company, or the number of media headlines it receives.  Machines can interpret this data of immense quantity with extraordinary speed, and using statistical regression methods, can identify - and appropriately incorporate correlations into its price predictions.

The price of a stock, at any given moment, can be thought of as analogous to a particle or object that has a bidirectional movement in one dimension (price), and a constant forward motion along the arrow of time, and this is important to why prices have long term dependencies, and why they possess underlying cyclical fractal shapes (self-similar repeating patterns). Betting on an up or down movement is not like flipping a coin at any moment - momentum is conserved to an extent, which causes stocks to run up or down.  Catalysts from the world can be viewed as nonlinear disturbances which can reflect that price... but on a deeper level, there is an important threshold between price and value that when not maintained, can prime the system for a correction toward greater equilibrium. Just as if you successively place pebbles onto a suspended piece of paper, eventually, the weight of the stones is too great for the paper to sustain its shape, or "morphology."  Interestingly, the most advanced of these quant funds is Renaissance Technologies, whose founder did pioneering work on the mathematics of algebraic topology.  This is important, because the machines read the price as a geometric object (a vector of price magnitude and direction) in an abstract topological space. The "paper" must achieve some morphism in under to accommodate the stress of so much "price energy" on the system.

These AI systems are not just one kind of architecture either - they are a network of varying types and fucntions of AI that cooperate to justify an action to buy or sell. These modular neural networks are very sensitive to sudden shifts in the importance of variables, because they are trained on enormous quantities of data (several terrabytes per day of training data at Renaissance).  That makes the machines extremely good at reading price by evaluating price as a time-series, and making predictions. However, chaotic events and historic shifts in technology, culture, and geopolitics are not factored by these machines, and catalysts can have more extreme effects as retail investors' react and prices move.  

Moreover, the introduction of increasingly concentrated wealth in automated trading architectures has caused what's called "overfitting" (in AI engineering terms) to a bias. Meaning that all the machines are interested at looking at the same types of variables, and combined with increasing wealth pouring into what are called Quant funds (quantitative), this overfitting has become more dramatic, pushing the momentum of stocks into a positive feedback loop - and the machines can push a price way beyond its value (or below it). 

At some point, a correction will occur, the S&P has acquired a tremendous amount of capital energy, with the Fed pouring out zero percent for the last decade and derivatives and ETFs soaring far higher than in the crash of 2008.

I predict a precipitous decline in the entire market, followed by (another) recession. During this time, the machines will be unaccustomed to a weaker, more balanced market - and will need to be re-trained. That means human-run hedgefunds may outperform the algorithms during the downturn.

I'm expecting a major catalyst, like a hack (Yahoo), or even a geopolitical event.  Even the most innocuous perturbation could cause a steep selloff by the algos, which will quickly snowball as Markets reel.  This is not an exaggeration. Algos control 70% of all institutional investing. Major institutions like Goldman Sachs, JP Morgan, Renaissance Technologies, and Blackrock are all moving to computerized trading, while simulatenously warning over the problems of overfitting.

Enter cryptocurrency - which itself is recovering from a massive shakeup to the record high Bitcoin, which had risen to near $3000 CAD, and the newer Ethereum, which also fell 25% last week.  his may indicate that algorithmic and trading volatility has become displaced beyond the confines of the market.  Crypto may not be the safe haven that it seems.

That being said - the S&P 500 always rallies after the inevitable bailout and quantitative easing. You could go long on the market and as long as you live long enough, make a decent profit of about 10% annually, averaged out. But it might be time to hold some cash, wait for the crash, then buy in those sectors where human investors are likely to go - where profits are. The crash will be done, when Amazon's RSI falls to the 30-40 range - that will mean a near 50% drop in Amazon's price to around $600.

I precict the time frame for this to occur will be sometime between now and the end of 2018.

For investors, defensive stocks are utilities, bonds, cash, and gold.

I am trading in the oil and energy sector, expecting a redistribution of capital after the crash, and an increase in oil prices to $60/bl by the end of 2018.

Of course, I don't hope this happens - I just think that it will.

No comments:

Post a Comment