Ulrich Benzings Blog

  …in my Weblog. I am a computer scientist (M. Sc.)  and certified Hedge Fund Advisor ebs/BAI. I work as a consultant in the areas of computer science, finance and quantitative trading strategies. If You would like to contact me, decipher the following code: ulrich(at) ulrichbenzing(dot) de.

You can find my profile here:  UHB - Competency Profile as PDF

And the technical white paper on my latest quant System here: The B- Quant - White Paper for Pros as PDF

The algotrader  is today long on Adidas, Metro, going long Arcelor Mittal, shorting Commerzbank, closing long position on Peugeot; Saint Gobain got stopped out.

The algotrader switched stocks (due to new stocks in the investment universe) and is now long Adidas, Metro, Peugeot, Fresenius Medical and ASML Holding while shorting Volkswagen.

The B-Quant is currently signaling to buy (long)  Adidas, Linde, Fresenius, Metro and Bayer and to short Volkswagen. Stop-Loss positions are about 5% below yesterdays close-values.

A huge amount of stock market indicators and trading approaches have been discussed in the recent decades. But for the most of them, their usefulness has never been empirically proved. Even worse, many traders concentrate on “ideal” trades without thinking too much about money management, a real portfolio, their long-term transaction costs, spreads, or slippage. But a single deal is, in the best case, an example for a good trading tactic. A good investment strategy needs more. These postings will outline my research and real-life results in a nutshell.

The maximum possible returns vary according to several parameters. Among others, these parameters are:
• The used investment and disinvestment algorithm
• The investment universe
• The money management approach
• Investment amount in relation to fees and liquidity
• Market Volatility

Conservative Stance:

Agressive Stance:

No Profit comes without risk. Therefore, we should not get carried away by these stellar results, but instead look at the associated risks. Many investors take the possible downturns as an important measure for the risks associated with a given investment strategy. The author has therefore included detailed information regarding the possible pullbacks in detail. We will start with a detailed look at the aggressive stance:

Statistics for 2001
StartValue 9.956.959,05
Total Yearly P/L 43,784%
-Intra-year Max was 15.969.149,26 at Date: 2001-09-21
-Intra-year Min was -11% at Date 2001-12-04
-Intrayear Pullback was not recovered intra-year.
EndValue 14.316.502,96
Total no. of Deals 489
Avg. P/L all Deals w/o fees 0,49%

Statistics for 2002
StartValue 14.176.759,91
Total Yearly P/L 111,589%
-Intra-year Max was 30.545.409,42 at Date: 2002-12-27
-Intra-year Min was -19% at Date 2002-10-16
-Pullback was recovered on: 2002-12-19
EndValue 29.996.535,28
Total no. of Deals 451
Avg. P/L all Deals w/o fees 1,101%

Statistics for 2003
StartValue 27.667.520,89
Total Yearly P/L 80,452%
-Intra-year Max was 50.179.132,99 at Date: 2003-12-30
-Intra-year Min was -10% at Date 2003-03-17
-Pullback was recovered on: 2003-04-02
EndValue 49.926.590,34
Total no. of Deals 574
Avg. P/L all Deals w/o fees 0,559%

Statistics for 2004
StartValue 50.408.461,76
Total Yearly P/L 15,544%
-Intra-year Max was 59.638.038,85 at Date: 2004-08-12
-Intra-year Min was -8% at Date 2004-11-01
-Intrayear Pullback was not recovered intra-year.
EndValue 58.243.725,00
Total no. of Deals 406
Avg. P/L all Deals w/o fees 0,213%

Statistics for 2005
StartValue 58.278.145,18
Total Yearly P/L 15,51%
-Intra-year Max was 67.723.853,65 at Date: 2005-12-29
-Intra-year Min was -5% at Date 2005-11-10
-Pullback was recovered on: 2005-12-01
EndValue 67.317.328,17
Total no. of Deals 455
Avg. P/L all Deals w/o fees 0,184%

Statistics for 2006
StartValue 67.548.353,20
Total Yearly P/L 20,539%
-Intra-year Max was 82.716.104,93 at Date: 2006-11-07
-Intra-year Min was -6% at Date 2006-07-19
-Pullback was recovered on: 2006-08-24
EndValue 81.422.253,42
Total no. of Deals 526
Avg. P/L all Deals w/o fees 0,207%

Statistics for 2007
StartValue 81.388.476,22
Total Yearly P/L 25,473%
-Intra-year Max was 108.135.031,31 at Date: 2007-11-21
-Intra-year Min was -6% at Date 2007-12-28
-Intrayear Pullback was not recovered intra-year.
EndValue 102.120.694,47
Total no. of Deals 508
Avg. P/L all Deals w/o fees 0,274%

Statistics for 2008
StartValue 102.193.350,81
Total Yearly P/L 223,913%
-Intra-year Max was 350.311.797,4 at Date: 2008-12-19
-Intra-year Min was -27% at Date 2008-11-10
-Pullback was recovered on: 2008-12-11
EndValue 331.017.866,33
Total no. of Deals 436
Avg. P/L all Deals w/o fees 1,744%

Statistics for 2009
StartValue 327.605.615,61
Total Yearly P/L 44,36%
-Intra-year Max was 480.049.506,95 at Date: 2009-04-27
-Intra-year Min was -11% at Date 2009-01-28
-Pullback was recovered on: 2009-02-27
EndValue 472.931.161,84
Total no. of Deals 174
Avg. P/L all Deals w/o fees 1,253%
We can see that we our smallest drawdown was five percent in year 2006 and our largest drawdown was -27 perc ent in year 2008 when following the aggressive stance. On the other hand, we reaped a total profit of 3000 percent in a little more than eight years. In real life, however, we could not have reaped this lever in this scenario because of the limited liquidity in european stock markets; but we could have
expanded into other market areas once we reached the natural liquidity barriers.

Statistics for 2001
StartValue 9.987.542,62
Total Yearly P/L 20,85%
-Intra-year Max was 13.759.796,01 at Date: 2001-09-21
-Intra-year Min was -13% at Date 2001-12-04
-Intrayear Pullback was not recovered intra-year.
EndValue 12.069.985,74
Total no. of Deals 680
Avg. P/L all Deals w/o fees 0,57%

Statistics for 2002
StartValue 12.031.660,33
Total Yearly P/L 57,061%
-Intra-year Max was 18.980.309,93 at Date: 2002-12-27
-Intra-year Min was -8% at Date 2002-08-28
-Pullback was recovered on: 2002-09-18
EndValue 18.897.051,18
Total no. of Deals 694
Avg. P/L all Deals w/o fees 1,314%

Statistics for 2003
StartValue 17.890.540,59
Total Yearly P/L 27,503%
-Intra-year Max was 22.843.203,63 at Date: 2003-12-22
-Intra-year Min was -7% at Date 2003-03-25
-Pullback was recovered on: 2003-09-03
EndValue 22.811.018,92
Total no. of Deals 789
Avg. P/L all Deals w/o fees 0,474%

Statistics for 2004
StartValue 22.867.715,46
Total Yearly P/L -0,173%
-Intra-year Max was 23.133.052,93 at Date: 2004-08-13
-Intra-year Min was -4% at Date 2004-05-04
-Pullback was recovered on: 2004-08-13
Statistic Data of Conservative Stance:
EndValue 22.828.187,48
Total no. of Deals 486
Avg. P/L all Deals w/o fees -0,065%
Statistics for 2005
StartValue 22.833.522,46
Total Yearly P/L 5,722%
-Intra-year Max was 24.172.639,94 at Date: 2005-12-29
-Intra-year Min was -1% at Date 2005-11-10
-Pullback was recovered on: 2005-11-21
EndValue 24.140.072,61
Total no. of Deals 515
Avg. P/L all Deals w/o fees 0,156%
Statistics for 2006
StartValue 24.165.164,55
Total Yearly P/L 6,883%
-Intra-year Max was 25.855.076,16 at Date: 2006-11-07
-Intra-year Min was -3% at Date 2006-07-19
-Pullback was recovered on: 2006-11-01
EndValue 25.828.475,59
Total no. of Deals 618
Avg. P/L all Deals w/o fees 0,17%
Statistics for 2007
StartValue 25.826.630,55
Total Yearly P/L 8,375%
-Intra-year Max was 28.346.916,05 at Date: 2007-11-21
-Intra-year Min was -2% at Date 2007-12-03
-Intrayear Pullback was not recovered intra-year.
EndValue 27.989.628,65
Total no. of Deals 614
Avg. P/L all Deals w/o fees 0,221%
Statistics for 2008
StartValue 27.997.159,11
Total Yearly P/L 41,456%
-Intra-year Max was 45.347.749,29 at Date: 2008-10-28
-Intra-year Min was -17% at Date 2008-11-26
-Intrayear Pullback was not recovered intra-year.
EndValue 39.603.629,77
Total no. of Deals 659
Avg. P/L all Deals w/o fees 1,084%
Statistics for 2009
StartValue 39.537.852,43
Total Yearly P/L 15,252%
-Intra-year Max was 46.507.076,93 at Date: 2009-03-06
-Intra-year Min was -6% at Date 2009-02-11
-Pullback was recovered on: 2009-03-02
EndValue 45.568.146,17
Total no. of Deals 294
Avg. P/L all Deals w/o fees 0,933%

We see that the drawdowns have eased further to -1% to -17%, while our profit has
now plunged to quadrupling the capital.

I did detailed paper-testing as well as real-world testing with several
banking accounts. It was possible to reproduce the results in practise, as long as
absolutely no deviations were made. In the real-world test, a profit of about 3,8
percent per month before transaction costs could be realized, which was in line with
theoretical results. It is however important that either inexpensive broker accounts
are used (like interactive brokers), or that the invested capital is large enough to
reach the capped brokerage pricing area.

Money Management is straightforward, not to say trivial. It is also the sole distin-
guishing factor between the three stances. The Conservative Stance was calculated
with up to 20 samesize positions. The neutral and aggressive stance used twelve
and six samesized positions, which is why this algorithm is to be considered non-
diversified.

We define Investable Assets as assets whose liquidity allows us to open or close
positions at will daily without changing the prices by our own transactions.
In practise, we decided to look at the 120 european stocks that have the highest
trading volume, assuming that we will be using non-diversified portfolio sizes of
6 to 20 assets and a total investment volume of up to 40 Million Euros. Certain
derivatives allow us to enter into large positions with minimized transaction costs
and no price movement caused by our own trades.

The slippage is precisely calculated. This means that if an asset opens below a given
stoploss mark, the trade is closed upon open and not on the SL mark (simulating
an unconditional SL). If the SL mark is crossed intra-day, we assume that slippage
is negliable and we close upon the SL mark.

Overadjustment is a severe problem for many quantitative approaches. Overadjust-
ment means that You have too many parameters that can adjust to the past data ,
therefore the system always models the past very good, but fails to correctly fore-
cast any future stock movement. In order to detect an overadjusting quant, one will
usually use the backward/forward calculation method.
A good example for overadjusting systems are many quants that use the percep-
tron/neural network approach. The first software version the author designed (in
1998) promptly ran into this issue. By using the Backward and Forward Calculation
approach, that issue could be solved.
A solution to this problem is to restrict the number of parameters that are adjusted
and to stick the others to certain fixed values… educated guesses. Another option
is to make this parameters sticky, ie not changing them during one or more runs,
but adjusting them between learning/evaluation runs.

It is imperative that automated trading approaches are evaluated against a long
timeframe in order to ensure that upward, downward, and sideward movements are
covered. Only after the successful test against such timeframes, real-world trading
may be started.

Dear Reader: here You can get my white paper for offline reading. get it here: B- Quant - White Paper for Pros.

For high-volume trading approaches, transaction costs are of extreme importance;
yet many traders do not take real transaction costs into account. It should be
obvious that a broker who offers either flat fees or at a maximum (cut) fee is best
suited for automated trading systems. Many brokers charge a base fee, a linear
fee and offer a cut. It is also of importance that partial executions should not be
charged extra. Last but not least, the broker of choice must offer the possibility of
shorting assets, an API for connection to the automated trading system, and the
possibility to trade at a variety of exchanges.
Our Trivial Transaction Cost Model consists of a base fee f1, a linear percentage fee
f2 and a cut at f3.
For example, Interactive Brokers offer all necessary features at a fee of f1=4,00 EUR,
f2 =0,1 percent, and f3=30,00 EUR.

Usually, one would not start with less than 5 million Euro. This is also the minimal starting point for a professionally managed SICAV.

My current quant system is usable on all markets, including the
commodities, currency, and bond markets. Therefore, there is virtually no practical
limit to the maximum size of assets under management. At the very least, several
billion Euro of assets under management should be possible without watering down
the possible profits too much. However, currency movements should be swapped
out when leaving the Euro-denominated markets.

Liquidity is paramount when using an automated approach. This is the reason why
the author concentrated on the most liquid european stocks. Although european
stocks are somewhat less capitalized than those of the U.S., the selected 120 stocks
are more than liquid enough to allow an easy investment and disinvestment without
influencing prices. Also, spreads in those 120 papers are usually minimal, and most
often are only 0,01 EUR, the lowest spread possible.

It is possible to produce high double digit percentage returns with artificial intel ligence-
based models on the stock markets, using a long/short equity approach. However,
a certain amount of assets is necessary to reach a costefficient ratio between assets
under management and cost of operations, usually at leaqst 5 million euros.