Title: Trading Systems and T3
1Trading Systems and T3
- Tim Tillson
- Denver Trading Group
- March 25, 2004
2Introduction
- Experience
- 40 years trading stocks
- 20 years trading options
- Serious work in technical analysis for 11 years
- Published in the Journal of Combinatorial Theory
and Technical Analysis of Stocks and Commodities - Proprietary C program as a test bed for filters,
systems and options research - Research partners
- Dave Chamness (Trading System designer)
- Roger Ison (Protein research scientist)
3General Market Perspective
- In a given timeframe, a market is either
- Trending
- Reversing
- A Random Walk
- A market will likely be doing different things in
different timeframes!
4Its the autocorrelation, stupid!
- How well does past behavior predict future
behavior?
5RWLK (NMS)
6RWLK Autocorrelation
7National Semiconductor
8My Favorite Trending Stock - RAMP
9RAMP Autocorrelation
1016 Period Sine Wave
11Sine Wave Autocorrelation
12A real trending stock (NYSE, Vol gt 100000)
13QQQ 2 day reverser!
14Equity on best QQQ system (In-sample, not AC
based)
15A really good reverserNYSE, Volume gt1000000
162 Year Equity Curve Reversing Stock
17Combination System
- Basket of 10 stocks which have strong positive
(trending) or negative (reversing)
autocorrelations - Trade each stock with a single adaptive
oscillator derived from the autocorrelations - Always long or short each stock
- Gives a degree of hedging
18Combo equity OOS since Jan 10 (50 days)
19Conclusions
- Long term trend following works on commodities
(Chamness, Fitschen) - Short term reversing works on stocks (Karnish,
Tillson) - Everyone spends all their time on the system.
WRONG! - Chose a good system and a timeframe, THEN find
markets which work with that system! - Indexes like QQQ are reversy, but harder than
selected stocks - Trade baskets of things that work for higher
Sharpe ratios and lower drawdowns! - If some constituent of a basket stops working,
replace it!
20T3
- Using Digital Filter Theory to build better
trading indicators
21Exponential Moving Average
22DEMA (Double EMA)
- DEMA E E(P-E)
- E E E(E)
- 2E E
2
Note E squared means Ema(Ema) not Ema Ema!
23(No Transcript)
24Problem DEMA overshoots
- Like to get in between EMA and DEMA
- Consider
- E(1v) E v
- If v0, we have an EMA
- If v1, we have DEMA
- V .7 is magic!!
2
Root Mean Square error is minimized when tested
over thousands of time series when v is
approximately .7
25TMA (Tims MA)
- TMA E1.7 E .7
- Use as a proxy for price - or -
- Use instead of EMA in systems
- TMA has only 30 as much lag as an EMA of the
same speed
2
26EMA, TMA, DEMA
27T3 Moving Average
3
- T3 TMA(TMA(TMA)) TMA
- Published in January 1999 issue of TASC
- T3 has less lag than an EMA of the same speed,
but is much smoother - T3 is not predictive, but is a really good
measurement tool - Definitely improves systems!
28Best Books (if youre really motivated)
- The Mathematica Book Wolfram
- Numerical Recipes in C Press
- Matlab Manuals The MathWorks
- Digital Filters (3rd edition) Hamming
- Adaptive Signal Processing Widrow
A multi-year relationship!
May have to read several times!
29Thanks for listening!!
- timtillson_at_comcast.net