B-Validus Presentation

Optimising Profits while managing Risk

Agenda Objectives of this presentation Common Trading Strategies and their shortcomings Modern Bayes: an introduction B-Validus: a Better Approach

Objectives Explore the shortcomings of common approaches to trading Explain how using modern Bayesian techniques can optimise profits while managing risks Encourage you to take a free trial of B-Validus

Common Trading Strategies Analyze (market intelligence, patterns, analytic tools): Establish inefficiencies Identify trends, expectations and associated parameters Start trading (heuristics and news) Algorithmic trading (mathematics and systems)

Simplification of Basics Select stop loss Go short/long if trend is down/up Sell/buy when trend changes or hits lower/upper limits or barriers Get out if hit stop loss, keep trading otherwise until achieve objectives.

Shortcomings - 1 Missed Opportunities: Not optimising over the whole trading “cycle” Decisions are almost entirely triggered by the current price Likelihood is high of hitting stop loss more often than profit objectives Incomplete risk management: stop loss only

Shortcomings - 2 Increasing reliance on speed: More frequent transactions Human error & emotions Management issues: Opaque process Policy enforcement and guidance Methods for managing risk

Other issues Most Traders use a similar approach: Higher peaks and deeper troughs Higher costs Favours high-speed responses e.g. “black boxes”

Bayesian statistics-1 Formal statistical methodology, developed 1970-, gradually superseding classical methods. Characterized by mechanisms to: combine diverse sources of information (data, human expertise, historical sources) to deliver superior results Becoming the method of choice for critical applications

Bayesian statistics-2 An approach to statistics in which estimates and forecasts are based on combining prior knowledge (distribution) with sample data. Bayesian statistics is not a branch of statistics, but a completely holistic modern re-invention of statistical methodology .

Bayesian statistics-3 Provides tools and techniques for all statistical problems. Every 20 th century statistical method has a 21 st century Bayesian equivalent, for example: Bayesian forecasting (replaces standard forecasting); Bayesian hypothesis testing, etc.

Bayesian statistics-4 Bayesian methods formally employ information available from non-data sources, e.g. expert judgment, past experience, historical data Couples with sample data

Seminal Bayesian references J.M. Bernardo and A.F.M. Smith, Bayesian Theory, 1994, Chichester: Wiley. A. Gelman et. al., Bayesian data analysis, 2004, Boca Raton: Chapman & Hall. M. Goldstein and D.A. Wooff, Bayes Linear Statistics: Theory and Methods, 2007, Chichester: Wiley.

Modern Bayes-1 For process X and observed data Y, the data we see are f(Y|X), i.e. the data are some function of a process The function is expressed in the language of uncertainty, i.e. probability. Our interest is in the process, X.

Modern Bayes-2 What we want to know is f(X|Y), i.e. what the data Y tells us about X. Bayes theorem tells us that f(X|Y)=[f(Y|X)f(X)]/f(Y) { We can ignore f(Y) in this discussion } f(X) is the prior distribution, found by taking into account other sources of information

Modern Bayes-3 f(X|Y) is the posterior distribution for X having seen Y. Broadly, it captures - in probability form – what we know about X having observed Y, and taking into account what we already knew from other sources f(future prices/trader’s belief, conf, prices to date) f(X|Y) provides our forecasts and error bounds

Bayesian: risk + attitude f(X|Y) combines naturally with utility functions (to quantify attitudes to risk) and with decision making (maximise expected utility -> Profit) In trader terms, f(X|Y) represents what you know having seen prices

Bayesian applications NASA – space shuttle live telemetry and propulsion systems Medical diagnostic systems Nuclear industry and high-complexity software testing Complex model validation methods (e.g. war games, reservoir engineering)

B-Validus A Better Approach Original Concept Developed 2001-2003 First prototype 2004 Shadow trials with traders 2006 2 nd trials 2007, 3 rd (Oct.) 2008 Version 1.0 March 2009

B-Validus – Design-1 Combine the trader’s know-how and skill with the power of modern mathematics & modern computers. Think longer term than just current price; optimise and manage risks over an entire trading cycle

B-Validus – Design-2 Continuously monitor actual prices, adjust accordingly and detect early signs of the market behaving unexpectedly. Quantify risk in real-time as actual prices impact a trader’s position

B-Validus: Real-time Risk Management-1 Customized “Static” Parameters: Maximum acceptable loss (stop loss) Risk attitude (utility for risk) Max short/long position Max buy/sell per trade Desired end position (long, short, or completely backed out)

B-Validus: Real-time Risk Management-2 Customized “Dynamic” Parameters: Trader’s belief in where the market will go (forecast) Trader’s confidence in forecast Expected versus calculated volatility

B-Validus: Real-time Guidance-1 Monitors actual prices to: Maximise realisable profit within the specified constraints Estimate risk for the remaining period taking into account: Trader’s forecast Stated confidence Actual price to-date versus forecast Volatility to-date

B-Validus: Real-time Guidance-2 Provides diagnostic alerts e.g.: Original price forecast requires revision state of market, e.g. market gone to a pure game Increasing risk of hitting stop loss Recommends the optimal trades to make within the specified risk parameters.

Who should use a Bayesian approach? Traders who have a view, however formed, valid or not, about how prices will evolve, then Bayes provides an advantage . Managers who wish to provide guidance, apply risk policies and/or improve trading transparency

B-Validus Free Trial Visit http://www.bayesline.com Register Request a trial license

Price Forecast & “ Actuals ”

Where do I get my forecast? Traders can supply such a forecast: when they have information reflecting the market expertise of the trader, or confidential info on future prices, or just a feeling for how the market will develop, or provided by technical analysis of recent prices, e.g. from a package

I have ideas about future prices, why is this useful? In short, Bayesian approach will achieve larger profits when your feelings are approx right, and smaller downsides when you are wrong Bayesian statistics will combine your expertise/beliefs with live data and so give better results.

More information, better handled If you think you know where the market is headed, Bayes captures that know-how and exploits it If you are right, the recommendations will help you make more money If you are wrong, you will receive early alerts You can review/modify your forecast at any time Commercial in Confidence – do not distribute

Commercial in Confidence – do not distribute

Decision-making Need to quantify uncertainties: but people don’t do it naturally. They need help from decision-support tools. One issue is that the mathematics is complex and high-dimensional Another is that problems need proper structuring But the Bayes approach can do this for you with your help

Managing risk Consequences of decisions must be made plain. Large losses are more consequential than large gains Most organizations pay premiums to avoid risks (insurance). This does apply also to trading A usual way to manage risk is to define a utility function for gains and losses

Utility functions Utility functions are generally concave, displaying risk aversion This means that large losses are considered not balanced by a large equivalent gain A $10m profit would be very welcome, but a $10m loss could break the trader

Insurance Premiums against Risks The premium paid is the difference in profits under different risk strategies The more risk averse strategy adopted, the higher the premium

Bayesian risk tool Bayes tools then can maximise realisable profits subject to the given risk attitude Also sets a stop loss

Diagnostics The model created by Bayesian methods has a forecast and an expected forecast error for every new price arriving Diagnostics compare incoming prices to model forecasts Unusual market behaviour is signalled by one or more diagnostics changing colour

Some Bayesian diagnostics To monitor overall forecasts as actual prices arrive Assess the impact of price deviations on your trading position Tell you when your forecast needs review and when a change is critical To warn of approaching Stop Loss

Decision-support tool The bottom line is that: if you work with a Bayesian approach, which embodies new advances in statistical technology and risk theory, and if you commit to the ethos that evaluating market uncertainties is crucial, You will be much better placed to exploit a market.

Decision Support Tool No tool can replace you: You remain the domain expert and decision-maker, but well-supported

Bayesian based algorithms Utility package to help determine your risk attitude. Expertise package to help elicit trader judgements. Forecasting models Bayesian updating: forecasts and uncertainties Optimisation algorithms

A Brief Summary A Bayesian approach: Helps the trader think about & structures his risk profile Manages risk Optimises profit over any selected trading interval within the chosen risk profile: Larger profits/smaller losses You can’t beat a Bayes approach and stay within risk! Provides guidance and warnings about how market develops

B-Validus Presentation

More Related Content

What's hot (20)

Viewers also liked (17)

Similar to B-Validus Presentation (20)

B-Validus Presentation

Editor's Notes