Tag Info

Hot answers tagged

9

The volatility in the indices long ago was similar in magnitude to what it is today. The problem you are seeing in your plots is one of compounding and scaling. Think of it this way- back in the mid 70's the magnitude of NASDAQ pricing was around \$100. Today it is on the order of \$4000, a change of 40x. In linear terms, a 1% change in the index today ...


3

As @babelproofreader mentioned, I recently blogged about the Roll model (see the original paper), which provides a very simple method for inferring the bid/ask spread based on trade prices. In short, you can estimate the cost using using the covariance: $c = \sqrt{\gamma_1}$. Where $\gamma_1$ is the $Cov(r_t, r_{t-1})$. (The R code is provided in my post). ...


2

ITCH does not disseminate any identifiers for the buy-side. They have a match number (used to correct or break trades) and a reference number (for displayed liquidity) and that's it. No other identifying features are present.


2

To construct best bid/ask from ITCH you must build a book incrementally from the messages in the data. Every message, except for system oriented messages, and non-displayed Trades, represent an order or an action on an order. Process the data, build a book, and you will naturally be left with the best bid/ask at the top of each side.


2

I was able to identify significant participants by order size on CME exchange. I think ITCH is even more informative that CME's data format. The trick is to learn very closely the incremental data and the order in which this data arrives. We can assume that exchange's Matching Engine and its market data distribution algorithm are programmed machines, ...


1

Always use a semi-logarithmic scale when looking at prices. It makes percentage moves of equal heights on your graphs.


1

SEC site showing Form 25 (delisting) filings from the last 4 years for all listing markets. As @user508 mentioned Nasdaq has then listed on their site here. NYSE lists them on their site here: www.nyse.com/regulation/delistings *Sorry for the lack of hyperlink for the last one. Only allowed 2 links due to lack of rep.


1

Regarding the second part of your question, - if you have relatively precise timestamps, you can use those to distinguish the cases you're interested. E.g. if one party took all three levels, the timestamps will be very close, or identical.


1

The cross-validation procedure does not turn on the choice of algorithm. Yes - calculate the prediction error of the fitted models when predicting the V'th part of the data. Combine the V estimates of prediction average using a simple average. Subsets should be randomly sampled (roughly equally sized). 2a. Subsets should not overlap. No. As long as the ...



Only top voted, non community-wiki answers of a minimum length are eligible