Feeds:
Posts
Comments
Just turn back the clock to before Bush invaded Iraq and draw a straight line from the state of America and the world then to now. Do you see how from the loftiest heights of optimistic future the entire discourse has descended into total sewage? America is being totally destroyed by these incompetent tribalists while they look for enemies all around the world.
Trump is a very very dangerous President because he’s a narrowminded ghetto guy with a gigantic ego and no knowledge of American or world history. He is incapable of reaching the intellectual level required to even think like a real President (such as Roosevelt). The best thing he can do is NOTHING for 4 years.

To make a sad comparison, I USED to think that Bush Jr. was the worst it could get, but Bush Jr. was intelligent enough to know his own limitations and surrounded himself with trusted advisors. Trump is intellectually far inferior to Bush (which is why he is a white supremacist) who has too much of an ego to listen to advisors. What the Hell sort of leader of the most powerful state in the world has news coming out about his inability to know what the issues are and all the news are about his temper tantrums and need to do what he wants and problems with making judgments about whether white supremacists should be embraced or not? He’s not emotionally developed enough for this role obviously. America is in extremely grave danger.

United States of America is in incredible danger from these racist fascists. Those who don’t embrace universal ideals will completely cripple the United States. I am sorry but without racial diversity and equality, America is nothing. America’s greatness is ONLY this and NOTHING else. Robert E. Lee is gone as is the Confederacy for a very good reason. They played no part in making America great. They were the failed experiments.

This is hopeful. At least the military recognizes how dangerous the neo-Nazi racists are for America. This stuff is cyanide for the American Empire. ALL INSTITUTIONS in public and private sector in America will unravel if this poisonous racism becomes legitimized.  This is poison that was stoked since early 2000s as racism was a key component of the military’s strategy in the Middle East.  This poison is much much worse than any of the flaky ‘blowback’ theories of Islamist terrorists.  All that is mostly orchestrated fakery along with 9/11.  THIS stuff — white supremacists, neo-Nazis, KKK, this is the real destructive force that can destroy the United States beyond any repair.  And don’t look at me.  I didn’t create this mess.

 

 

 

http://www.politico.com/story/2017/08/16/military-chiefs-comments-trump-charlottesville-racism-241714

It’s interesting now to return to one of the catalysts for this brand of racism and bigotry: the Zionist right who have been playing with fire by courting these groups for wars they wanted in the Middle East. These crazies are infinitely more dangerous for America’s survival than any Islamic terrorist. Islamic terrorism was mostly orchestrated by the powers that be: neocon Zionists and now they have created a nasty poisonous monster. These groups are the barbarians at the gate of Rome, not the stupid Islamists who probably are too uneducated to find America on a map.
TheRealNews has a great deal of coverage of how the Zionist right had been courting the far right in Europe and America in order to be seen as the frontier against Islam etc.  I won’t go into the details because the folks at TRNN had already done this, for instance this is from 2012.  Of course one could say that this is a stretch to hold the Zionists responsible for this strain of disease, but in fact they were the major players since Bush administration.  This problem is infinitely more deadly for America than all the Islamist terrorists in the Middle East.  Those guys are absorbed with their own problems and do not pose a fatal threat to the United States.  I mean, come on, really … while these guys are actually capable of completely destroying America.  These guys could not gain serious traction without the wars that Zionists wanted for whatever reason.  Those wars are total waste of American energy and resources.  Oil whatever.

These types of movements are a sign of disease in the American Empire.  Imperial power is held by the elite aristocracy when an Empire is healthy.  The Nazis tried the race-based power theories and they were destroyed.  Racial supremacy power structures are extremely primitive and don’t work very well.  The caste system in India failed miserably in the end and put that country in bondage to England partly because of this weakness.  United States of America became a global empire by 1945 not led by these racists but being able to absorb a diversity of industrious people who accepted the ideals of America and struggled over generations to find their place etc.  Trump’s ‘Make America Great Again’ campaign is unfortunately tainted by the KKK and neo-Nazi white supremacists and it’s not ‘very dangerous’ etc. but completely fatal to the survival of American Empire.  American wealth cannot be sustained by pushing these sorts of ideologies.  I mean, the white supremacists have not really done anything for anyone to accept their deluded ideas.  I’ve wasted some time chit-chatting with them on Facebook.  I have no idea what Trump thinks he’s going to do with this but I can assure him that American greatness is not due to these inwardlooking tribalists.  This type of ideology can only become prominent as a sign of deep disease in American Empire.  An Empire cannot survive indefinitely without health in its rulers.  This ideology will rip apart the weak cohesion gained in American economy by embracing anti-racist ideals.  The peak of America’s greatness in world power came through destroying the Nazis.  I don’t know what Planet Trump comes from, but here on Earth, it is a devastatingly bad idea to give any quarter to racist ideologies.  America will most likely not survive the economic dominance of China even with great efforts in the long run, but destroying the political and social equilibrium by supporting the KKK from the White House would make Trump the Trojan Horse of the Fall of Empire.

Paul Tudor Jones has moved into AI hedge fund direction indicating heavy momentum in the direction of the hedge fund business as a whole.  So it is a good time to ask: what’s wrong with the AI hedge fund models in terms of alpha generation?  I think the issue is that ‘large data analytics’ rather than algorithmic trading strategies is what’s wrong.  My own recent experiences with Machine Learning have been quite successful.  I’ve been doing signal processing + regression learning models to predict medium frequency price changes for currencies and commodities.  Here are sample results for grains.

 

corn-5m-5000soybeans-5m-5000wheat-5m-5000

These unpolished results may not look so good but the Sharpe ratios show significant alpha in deep markets.  So my view is biased.  I personally do not believe that the largeness of the dataset is the fundamental driver for alpha but rather a better understanding of the PREDICTABILITY OF PRICE CHANGES.  The latter focus is to restrict attention on signal processing that allows AI to be able to detect information rather than artificially creating a haystack in which to find a needle and thereby force sophisticated AI algorithms to overfit to noise.  I downloaded a sample dataset from Numerai today and at least for algorithms that I feel are pretty good (and able to deliver the results above and much better ones for currencies and energy) such as glmnet and ksvm had no better performance for the Numerai dataset than misclassification 0.492 on test sets.  I’d like to generalize this issue:  AI and algorithmic trading is clearly the future of alpha generating business but the substance of where alpha is to be found is badly addressed by ‘large data’ for the sake of largeness.  This is the bias of machine learning that I had fallen prey to myself years ago.  Alpha resides in understanding issues of price movements that are not automatically detected by machine learning algorithms.  One must carefully construct the features from signal processing for machine learning that produces results.  I believe at the moment that a key component of this is a return to Mandelbrot and ‘multifractal’ price series and I use wavelet analysis to produce features.  In fact, I believe right now that the future of hedge funds, i.e. organizations that can outlast the current generation of hedge funds that will struggle to produce returns and flounder around attempting to gather quant armies etc. is to quickly get away from the largeness of the data and focus on organizing fundamental research on price predictability via signal processing.  Second and third order improvements will reveal themselves only when the understanding of price change predictability reaches the same level of refinement as the basic set of machine learning algorithms that are widely understood today and occur in bestselling textboks and O’Reilly books.

In a sense, the translation of what technical traders had done since the 1920s now needs to be understood with a proper respect for the intuition of market behaviour and translated into higher frequencies for the fundamental component of alpha.  Companies that can organize themselves to do this in a calm deliberate manner without being pushed and shoved by the fads will probably inherit the entire industry.

 

zulfikar.ahmed@gmail.com <zulfikar.ahmed@gmail.com>

Attachments7:08 AM (56 minutes ago)
to harrington, jharris, jhp, jhricko_4, jianjunp, jinha, jjbrehm, jlind, jlondon, jlw, jmateo, jmerseth, jmetcalf, jmg, jmogel, joel, joelms, john.aldrich, john.beatty, johncrawford53, jose.oliveira, josesoto, josue, jpadgett, jpbalz
Ladies and Gentlemen,

Academics love the random walk model deeply.  Traders and hedge fund managers obviously don’t.  The former erected a gigantic edifice based on unpredictability of returns based on equivalent martingale measures; the entire industry of option pricing including my own rely on price process being a martingale.  On the other hand, hedge fund managers and chartists, not to mention the latest AI systematic hedge funds exist by convincing investors that they are clever and can find systematic anomalies.

What is my twist to this old debate?  I will claim that in medium frequency, 1 minute to 15 minutes, liquid assets such as currency and commodities have predictable price changes almost uniformly.  Rather than merely showing you statistical tests only, I will produce systematic strategies that are profitable net of transaction costs that are based purely on future price change predictions without any need for anything other than price data.  This will demonstrate two things: (a) liquid (currency/futures) markets today are inefficient at 1-15 minute time scales, (b) in order to make these markets efficient one must invest capital to force them to the random walk conditions of economic theory.

Without further ado:

The key tool I employ is a multiscale decomposition of PRICE series by a discrete wavelet transform and extract from this a set of pseudo price series at every level.  For these pseudo price series at different scales, we construct a set of technical indicators.  Then we use the changes in these technical indicators as the predictors for pseudo-price changes.  Then we combine the predictions appropriately to form an estimate of the price changes of our original series.

The above procedure may seem arbitrary; however, it is the adaptation of two sources of ideas.  On the one hand chartist traders have an art and sensibility for patterns in prices that have an encyclopedic set of rules of thumbs that are attempting to disentangle behaviour of prices at multiple scales (e.g. take a short position in XYZ if RSI says overbought but not if the underlying trend is bullish).  A second source is Benoit Mandelbrot’s original focus on ‘multifractal’ prices.  The above algorithm attempts to make precise using well established wavelet analysis tools.  An actual implementation (confidential) is attached for your evaluation.  The code is meant for backtesting a strategy that takes position based on the sign of the prediction as entry point and exits with a stoploss at 2*sigma and takes profits at sigma in period (1m,5m,15m) training on a lookback of 1024 or 512 points.

The machine learning algorithm I use is glmnet although I don’t think that the particular ML algorithm matters; it is possible to print a performance metric for each fit such as MSE for price change prediction.

What makes this work worth sharing is that the strategy seems to perform well across many currency pairs and commodities although not for all frequencies.  For example, for crude oil, 15m produces an unusable equity curve while 5m produces an implementation-worthy strategy.

I am attaching the equity curves of the backtests on 4000 points in the last month, currencies in 1m and commodities in 5m/15m.  Visual examination should convince you that these are implementable and typically have 6+ annualized Sharpe even after transaction costs.  The relative universality of these strategies suggest that indeed we are getting predictability of price changes at these scales, which is an issue which is of interest in finance as a science.  Of couse at 5-15 microseconds, HFT firms detect and profit from predictability.

In conclusion, it is true that markets are inefficient as the traders and hedge fund managers believe, and more specifically the inefficiency that I have discovered is not a little anomaly for any secret knowledge (for example, on rainy days traders are lazy and so they create an anomalous pattern); rather, I am providing an entire time scale for which all the liquid high volume markets have a statistical arbitrage opportunity.  So the efficient market theorists are wrong because their perfect world is sort of like a socialist utopia: they would be right only if a great deal of investments are made to remove these large anomalies from the markets but not ‘natural’.  The efficient markets only occur with funding and work by traders and funds,  On the other hands, the chartists are right about price patterns leading to predictability but one requires a precise tool to extract the predictability.  (I claim that wavelet decomposition is the most natural tool for this).  Technical indicators must be applied to each scale separately to obtain good price predictions by any regression learning algorithm (the difference between using machine learning algorithm say glmnet and kernel support vector machines is minor).

This is a big result because the markets are liquid high capacity markets which suggests that these results naturally demand systematic hedge fund with institutional capital to remove the statistical arbitrage at an entire time scale.  But the importance is also that they show that mining giant complicated datasets is probably looking for overfitting to noise issues without any insult to the AI funds that deploy large data predictor sets.  Predictability of price changes is much more clear and straightforward in currencies and commodities for certain and perhaps in equities as well.  It is not the AI algorithm that is important in my work but a concrete realization of examining prices simultaneously at multiple time scales in a coherent manner, that is, the signal processing aspect.

Thank you.

11 Attachments