Date post: | 11-May-2015 |
Category: |
Business |
Upload: | george-tziralis |
View: | 1,562 times |
Download: | 1 times |
Detecting Important Events using Prediction Markets, Text Mining, and Volatility Modeling
George Tziralis & Panos Ipeirotis
markets and efficiency
• strong: prices reflect all information, public and private
• semi-strong: prices reflect all publicly available information
• weak: past prices cannot be used for market prediction
prediction markets?
• Perhaps, the relationship between price and information is no more clear than in prediction markets (Pennock et al. 2002)
• let’s start from the weak form of efficiency
• is it possible to predict future prices using only past ones?
playground
• InTrade
• political contracts
• US nominee elections 08
• 800 - 1000 time series instances for each contract
machine learning
• various algos, potential inputs etc
• lots of both trials and errors
• finally selected Support Vector Machines
AI for PM prediction
• poor results
• very close to / worse than random walk
• suggesting weak efficiency
prices, diffs, returns
• low autocorrelation
prices, diffs, returns
• low autocorrelation
• garch analysis: ARMAX(0,0,0) for the mean
• TS ~ noise
• evidence of weak efficiency
variance
• typically stable through time
• not in our case
• GARCH(1,1) !!
• autocorrelation
• volatility clustering
volatility clustering
• large price changes are followed by similarly large changes in price, of either sign, and its opposite
• typical in all financial markets
• not yet utilized appropriately for event detection
• big price changes in periods of high volatility should not suggest significant events
event detection
• the task of monitoring news corpus to discover stories that discuss a previously unidentified event
• the vision of a robust system that would monitor news streams and alert on events
• principal assumption: information organization by event, rather than by subject
event detection, so far
• correlate a simple change in market prices or returns with a signal of an event
• naive approach, assumption that volatility remains stable
• this is not the case
volatility modeling
• introduce a GARCH model for the variance of prices
• concurrent analysis of volatility and text
• identify important events more robustly
text mining
• various options, lots of experiments, too
• let’s try a naive one here
• data from google trends
• keywords: nominee’s name
events affecting volatility
• unanticipated events
events affecting volatility
• unanticipated events
• introduce unexpected new information in the market and
• increase volatility after they become known
events affecting volatility
• anticipated events
events affecting volatility
• anticipated events
• increase volatility the days before they happen,
• introduce new information when they happen and then
• decrease the volatility allowing the market to stabilize in a new “equilibrium”
discussion
• prediction markets + volatility modeling + text mining ➜ strong potential for event detection
• work in progress, need ur feedback
• various future research directions
thank you!