I have a collection of crypto stat/arb strategies I plan to trade as a portfolio of strategies. Each strategy trades a small mean-reversion portfolio of loosely cointegrated coins, based on a bayesian state-based model. The returns in cryptos for this sort of strategy are phenomenal, however,

We have been working on some option strategies and wanted to get a sense of how well BTC and ETH options are priced on Deribit, i.e. is there a substantial IV premium over realized volatility or are options fairly priced. At first glance, based on the documentation, it seemed that Deribit options

In the previous posts I described an Reinforcement Learning approach to “Learning the Exit” part 1, part 2. My initial conclusions there have been: reward smoothing (with the labeler) leads to more robust results than a reward on position exit without smoothing the learning process struggled and

As described in my prior post Learning the Exit (part 1), I have a model that indicates mean reversion entries with ~81% accuracy, however I did not have a good approach in handling the exit. While 81% of MR signals had a minimum profit of 25% (of prior amplitude), the mean profit available was

In the prior post I showed results for a new equities strategy which uses a combination of signals to create and risk manage a high-momentum portfolio. Further investigation revealed that I had neglected on a couple of fronts: failed to account for dividends (which are substantial) some data issues

I have been thinking about the recent institutional buying that has propelled the price of bitcoin to stratospheric levels; in particular considering how one might detect some of this interest early. Bitcoin and crypto in general is quite interesting in that at some level, due to the decentralized

I have a couple of equities strategies that I will start trading shortly, and I want to understand the risk from all angles. Towards this end I try to utilize both market signals and exogenous unstructured data to minimize surprise and maximize selection or prediction efficiency. In thinking about

In the past have been a HFT market maker for FX and other traditional instruments, however have not investigated exchange-based market making in Crypto. As I have proprietary signals applicable for Crypto, thought it would be worthwhile to investigate the difficulty of market making on crypto

Most signals I deal with are noisy, reflecting noise of underlying prices, volume, vol of vol, etc. Many traditional strategies built on such indicators might either: use signal to scale into position such approaches have to deal with noise to avoid thrashing, adjusting position up and down with

In the prior two posts, investigated: Subspace Projections: feature selection (1/3) Information Geometric: feature selection (2/3) In this post will evaluate feature importance as implemented by Random Forest and compare to Information Geometric approaches. Here is an outline of what would like to

As mentioned in the prior discussion feature selection (1/3), of primary interest is understanding the contribution of each feature in x⃗ to the outcome or class labeling function f(x⃗ ) . One way to examine this is to understand how the distributions: p(xf) , the probability distribution of

I am often confronted with the problem of trying to reduce a high dimensional feature set to a, smaller, more effective one. Reducing dimension is important for machine learning models as: the volume of the “search space” grows exponentially at a minimum rate of 2d for binary categorical

It is fairly easy to recognize price momentum with price-based indicators ex-post or with lag. Price based momentum signals tend to have lag issues in recognizing the start and end of a price move as there is a tradeoff between noise and lag [1] that can’t be defeated without future information

Following on from the prior post, want to discuss the problem of sample independence. Many machine learning models in finance deal with timeseries data, where samples used in training may be close together in time and not be independent of one another. There are very few features in finance that do

I have used machine learning in trading strategies over the past 10 years. However my use of ML has often played a relatively small role in the overall design and success of the strategies. I use ML in specific signals or strategy sub-problems where the data / problem setup tends to have a robust

There are times when need to label a time series, identifying periods of momentum, trend, mean-reversion, etc. Directionaly labeling timeseries has a wide variety of applications: labels can be used for supervised learning analysis of microstructure around larger price moves conditional analysis

In the prior post Information In Volatility Structure [1] applied the SABR model to fit noisy raw option price data of approximatelty 700 million prices across a 10 year history of 2700 stocks. The point was to examine a hypothesis: does supply / demand imbalance in the options market express in

With market making we can try to be neutral by skewing prices in such a way as to maintain a neutral position. To the extent that the market can become 1-sided (in momentum) or may have large sized requests (if offering at different sizes), one’s portfolio may require explicit hedging. In a live

Investors are often looking for uncorrelated returns so as to better diversify. If one looks at world indices & equities, there is much less diversity between assets than there was a decade ago, indeed the cross-market correlations are remarkably high. On the other hand, from a trading

I have 4 Bitcoin L3 exchange feeds running smoothly out of a data center in California (which is slightly closer to Asian exchanges and Coinbase than the east coast). It took a bit of error handling and exponential back-off, to handle the unreliability of connectivity with these exchanges, where

I have implemented 4 bitcoin exchange interfaces now that produce a live L3 stream of orderbook updates + trades of the form: Screen Shot 2015-02-10 at 6.33.12 PM Given the above, can reconstitute the orderbook as it moves through time, and can likewise be used to creat

It seems like every other month there is a new bitcoin exchange. For the purposes of trading research & backtesting it is important to have historical data across, at least, the most liquid exchanges. My list would be at least: USD/BTC bitfinex (15%) bitstamp (5%) coinbase (new, but likely to