|
||
| Investor/RT Tutorials Backtest Tutorial by F_Trader (User Submitted) Revision 1.00 / February 26, 2003 |
Click here to download Word (.doc) Document
INTRODUCTION:
These pages are intended to help IRT users making their debut with the
backtesting feature. Basic knowledge on RTL language and IRT syntax for Trading
Signals and custom indicators will of course help. Anyhow, this document was
not written to replace the current Linnsoft documentation that you may find there:
http://www.linnsoft.com/tour/tradingSystems.htm
http://www.linnsoft.com/tutorials/ruleTesting.htm
but is mainly intended to bridge
the GAP between :
ð
how one manage, in his realtime daytrading activities,
the signals which can be generated by his own indicators on his charts
AND
ð
the best way to build
a full trading systems from these indicators (and implement money management
rules) with the very powerful tools available in the RTL programming language
and the IRT backtesting features.
Finally, being a mainly E-minis trader located in
- the main
- all trading systems examples are based on a 5 minute
timeframe, the default trading size being two ES contracts.
Of course, all explanations can be easily adapte
I really hope you will enjoy this Tutorial which gathers together most
of my backtesting experience with IRT for the past year.
All scripts here should be run with IRT version 5.9.7 or later.
Do not hesitate to contact me for question on this paper, or for
reporting possible tipping errors or inconsistencies by email at f_trader02@yahoo.fr, using the Linnsoft
Yahoo group or thru the #Linnsoft chatroom on the OTHERNET IRC server.
Happy Backtesting
Eddy
PS: please consider this very first version as a DRAFT / Check for any
new revision of this document which will be most likely advertised on the
Yahoo group or on www.linnsoft.com
_ACTUAL
TABLE OF CONTENT__
CHAPTER 1: Basic guidelines
for writing entry and exit trading signals
1.1)
Distinction between End Of
Bar (EOB) and Intra-Bar (INB) signals
1.2)
Market order type and slippage
considerations for EOB and INB signals
1.3)
Adding a “filter” to our
entry rules
1.4)
Managing reverse rules
1.5)
Tips for getting EOB /
INB signals markers on your chart and
about the equity curve charts
1.6)
Detailed scripts of the trading systems
discussed in Chapter 1
CHAPTER 2: How to add Money
Management rules to an IRT trading system
2.1) Using an EOB
approach for money management rules: a logical choice
2.2) Definition of
the Money Management V# variables.
2.3) The Money Management (MM) rules explained
step by step
2.4) How to simplify the “maintenance” works
around backtest or how to use exactly the same single trading system for
testing different strategies
2.5) Detailed scripts of the trading systems discussed
in Chapter 2
_VIRTUAL
ADDITIONAL TABLE OF CONTENT__
For information only, here are the chapters, NOT yet on paper, but which could get out of
my head one day an
3) Custom reporting with the NOTE Token
This NOTE token allows you to show any V# variable value in the detailed report file: you may use it to build a fully customize report
4) Getting your backtest rules in realtime on your
charts
or how to get live on your charts you’re the same
entry/exit/MM signals than the one you get in your backtest (without running
any backtest ..)
5) Managing multi-periodicity (and even
multi-instrument) signals
This issue is already discussed a bit in section 1.5
6) Optimisation of your trading system parameters
IRT allows you to automatically
fin
7) Max Favorable / Adversary
excursion and Intra-trade Drawdown analysis
Detailed MFE/MAE analysis and reporting for all backtested trades can be achieved using some V# variables
CHAPTER 1 : Basic
guidelines for writing entry and exit trading signals
We will start explaining here the distinction between:
A) an
“End Of Bar” (or EOB) Signal
B) an
“Intra-Bar” (or INB) Signal
and the impact of this essential
choice on the way you may write ALL your trading rules whatever their type : i.e.
entry signals, filters, profit taking rules, stop loss rules, etc
In fact, when designing a trading system, the first step is to check
that you have a coherent approach between:
ð
your trading style, i.e. how you manage your signals
in realtime
ð
and the way you intend to transcript them for backtesting
purpose
In order to write coherent trading rules and not cheat with your
backtest (ie enter a signal at a given time using
“future” data), it is essential that you understand all nuances between EOB and
INB signals. By experience, I guarantee you that the main pitfall
before getting “realistic” backtest results are here...
You will find that it might be interesting to open twice this document
so that you may follow at the same time how the systems are building, and how
the final result (ie IRT text transcript of all the
rules) looks like. These transcripts are located at the end of each chapter.
1.1) the
distinctions between EOB and INB signal
We will consider in this chapter a basic breakout system based on 5
minute bar chart.
In this system, you enter a long position when there is a breakout of
the Highest High of the 10 previous bars, and you exit it at the break
of the Lowest Low for the 7 previous bars..
A) End of Bar Signal (EOB)
If your system is : wait for the first CLOSE of a given bar to be
higher than the highest high of the 10 preceding bars to go long,
then your entry rule will have following logic :
If CL > STAT.1 Then BUY 2 at Close Price
with STAT
= Maximum Close for Preceding 10 Bars
Of course, you need to use STAT.1 in this trading signal in order to
simulate the highest high of the 10 PREVIOUS bar and exclude the high of the
current tested bar
Note: Since version 5.9.7, you could also use the syntax: CL > MAX
(HI.1, 10)
In this case, what is important is that you wait UNTIL the end of the 5 min
bar to make a decision about entering a trade, because your rule is : wait that
the CLOSE of the current bar breaks the highest high of the 10 previous bar (by
the way, this particular value is known since the start of the current 5 minute
bar)
ie with EOB signals, you don’t
care at all about what the market could do during 5 minute : there could be
1000’s of trades (which will be compacted into the Open / Low and High
information), you will be using only ONE trade data to take a decision : the
CLOSE of the bar
The corresponding trading system, incl. Filter rules (as explained later
in section 1.3) will be named REF_SYSTEM_A_EOB and its scripts, as other
in this chapter is available in section 1.6.1)
B) Intra-Bar Signal (INB)
In this case, your entry signal can be described as:
As soon as there is one trade done at a value higher than the previous
10 bars high, you enter the trade.
While trading in realtime with this approach, this requires you (or
trading platform) monitor all the trades as they might appear in a “Time and
sales” windows for example, because you need only one tick i.e. one trade with
the adequate price to trigger your entry.
Anyhow, in the backtest, your entry price will be the Highest High of
the previous 10 bar. This value is not listed by default in the “Rule Price”
window, so you need to store it using a V variable.
In IRT syntax, this will be done thru a SET statement like SET(V#2,STAT.1) which will appear in a Trading
signal at the top of the rules list (the STAT token has the same setting than
in A))
We will name it Store_Bkout_value and
it will be associated to a NONE rule (i.e. "no action" rule)
Please note that you will need to rename the STAT token in the Trading
signals definition because he will be used several time.
This way, the value will be updated at every backtest
iteration, i.e. for every new 5 minute bar.
The entry
trading signal associated to the BUY action will be then:
If HI > V#2 Then BUY 2 at Value V#2 Price
For more: see the script of the system named REF_SYSTEM_A_INB in
1.6.2)
Note:
Do not forget to go to the “Set Up / Preferences / User variables” menu
to give a title to this V variable.
1.2) Market order type and slippage consideration for EOB and INB
signals
OK, this part is not fully linked to IRT issue but it helps understand
the difference between EOB and INB signals
A) End of Bar Signal
In such a system, you do not know “in advance” what will be your entry
price as long as the 5 minute bar is completed (of course, you do try to enter
on the CLOSE of the bar)
So basically, if your system trigger when the CL data is printed, you
would enter the market as soon as possible, typically with a market order,
i.e. you will experience slippage. Of course, this impact of the slippage
(let’s say one or two ticks for futures) on the
Indeed, it might take you between 5 and 30 seconds to send you order
thru your broker platform, not so an issue if you are working with 30 minute
bars, but this is a big issue if your are working with 1 minute bars…
B) Intra-Bar Signal
In this
case, you know “in advance” (i.e. as soon as a 5 minute period starts) what
could be your entry price for both a long or short trade. But, at the opposite
of EOB signals, the signal could trigger (in realtime) at any time, not only at
the completion of the 5 minute bars. But knowing in advance the possible entry
price allows you to use pre-programmed limit orders (for both long and
short entry): i.e.
you
will be updating the corresponding values of the resting go long/short orders
every 5 minutes, at the start of every new bar… Such an operation could take
few seconds, so in this case:
ð either you may experienced slippage is the
breakout occur just at the start of the bar (because you don’t have updated yet
the value of your resting limit order)
ð either, and this is the case most of the
time, you won’t experienced any slippage at all, i.e. your limit order will be
filled at the exact price
C) The
bid/ask story or how to make sure that the “backtest fill” is a realistic one
Personally, I trade the S&P E mini which always have (during the day
session) a single tick (i.e. 0.25 point) difference between the ask and the
bid.
However, having a bid/ask at 835 / 835.25 and a buy limit order at
835.25 do NOT guarantee you a fill.
This is why I would write, in such a case, my entry rule as
HI > STAT.1
and NOT as HI > = STAT.1
Indeed, I want
to make sure that the limit order (I simulated in the backtest) would be
filled.
There will be
more example of this conservative approach later in this tutorial. The main
idea is that I prefer to have a pessimistic backtest result than an optimistic
one…
1.3) Adding a “filter” to
our entry rules
Adding Filter provides a very good illustration to the difference
between EOB and INB signals. Let’s consider the typical ADX filter, i.e. a
breakout will be only traded if the ADX is above 25 AND the ADX is increasing
A) For EOB rules
In such system, you enter a trade as soon as the signal is triggered, i.e.
when the 5 minute bar is completed. So you may use all the price data of the
current bar in your filter calculation, including of course the CLOSE
The entry signal for a long signal would be then:
If ( CL > STAT.1 AND
ADX > 25 AND ADX > ADX.1) Then BUY 2 at Close Price
B) Intra-Bar rules
In this case, as your signal may trigger as soon as the new bar starts
to form, it is not possible to use the current close in the filter calculation
If this is not clear yet in your mind, remember that the IRT backtesting
makes calculation ONCE at the end of every time frame (here 5 minutes), so if
you simulate an entry during the current bar, you need to limit your data until
the previous Bar data (i.e. CL.1 or ADX.1), so that IRT does not use the CLOSE
which will historically be printed after your possible trade entry time.
This is why the entry signal for a long trade would be then be here:
If (HI > V#2 and ADX.1 > 25 and ADX.1 > ADX.2) Then BUY 2 at
Value V#2 Price
(with V#2 being set as STAT.1)
WARNING: The rules: If (HI >
V#2 and ADX > 25) Then BUY 2 at Value V#2 Price is clearly incorrect because
“future” data is used to take a decision in the past.
By the way, IRT or (any other trading software) doesn't have sufficient
logic to warn the user with “an error message”. So always verify the rules
logic behind any backtesting report producing some extraordinary results…
Note:
For INB and EOB, I have used exactly the same filter rule for both LNG
and SHT sides, so I could have basically have used a unique trading signal name
: of course, there are usually different, so I'll keep naming them __LNG_FILTER
and __SHT_FILTER
C) Adding Filter as CONFIRM rules
As soon as your trading system starts to be a bit complex, I would
recommend to have a modular approach, i.e. for example, you may include your
filter with a CONFIRM rules, just before the actual BUY or SELL actions in the
list of trading rules.
This is the way the script features at the end of this chapter are
structured.
Indeed, such CONFIRM rules will greatly improve:
1) the legibility of
the structure of the trading rules
2) the execution times
of your backtest (as the possibly complex entry rules calculation will be only
done IF the filter rules returns true)
3) allow you to use
the same filter rules in different trading systems
Please note that the most common filter will be in fact a TIME FILTER
Let’s say that you will not enter trade on the first or last 10 minutes
of the daysession, you could write it in a Trading signal as: (TIME > 1540)
AND (TIME<2150)
You will notice that I do prefer write this rule as: (TIME > 1540)*(TIME<2150) > 0
because it look like such approach
results in quicker calculation
For example if you want to enter a trade only on the first or last hour,
you may write the corresponding signals as:
(TIME > 1530 AND TIME<1630)
OR (TIME > 2100 AND TIME <
2200)
or as : ( (TIME > 1530)*(TIME<1630) + (TIME > 2100)*(TIME < 2200) ) > 0
PS : As I never hold any position overnight, all my
trading systems will include a Market close rule (MKT_CL), i.e. exit all
position when the trading signal TIME=2200 returns true…
Do not forget to use such Time filter if you have 24 hours data
available (like it is the case for ES), otherwise you will end up triggering
trade at 3 in the night: indeed the IRT backtesting data will be based on the
default session of the instrument tested
1.4) Managing
reverse rules
OK, before
investigating the implementation of money management rules, we need to see how
to manage reverse rules, which can be a bit tricky when combined with filter rules…
You might think: reverse rules are easy to master!
I will answer
that this might be the second major sources of mistakes after the confusion
between EOB and INB signals.
By the way,
please note that following comments are fully vali
A)
With
no dedicated f ilters rules
Let’s start with a system having “NO filter rules” and a set of
symmetrical rules: i.e. the exit rules of a Short trade ALWAYS trigger the
start of a long position
In our example, this would mean that breakout perio
CL < STAT.1 but with STAT = Minimum of the Low of 10 preceding bars
(instead of 7 bars)
In this case, implementing the reverse rules is quite an easy task:
For exiting a long position, you just replace SELL by a REVERSE-SELL
& SELL SHORT action
For exiting a short position, you just replace COVERSHORT by a
REVERSE-COVERSHORT & BUY action
So the order of the rules could be as follow:
BUY rule
SELL rule
REVERSE-SELL & SELL SHORT
REVERSE-COVERSHORT & BUY
action
In IRT, these REVERSE rules allow to Exit a long position and Enter a short position on the same bar! We will tell more on
that later, in section 2.3
B) With a
filter rule prior to the entry signal
Let’s consider now our breakout system having the same 10 bar period,
BUT with some a filter for initiating a NEW position (short or Long)
Here it is just a bit trickier. Indeed, in this case, when you get your
signal for exiting a Long Position, you must make check the “filter” status to
know if this breakout of the 10 previous bar results in:
ð
Exiting a Long position and remaining flat (filter
status = false)
ð
Exiting a Long position and entering a Short (filter
status = true)
In this case, you just need to write in the following order (let’s
consider first only the LONG side rules before mixing both SHORT and LONG positions)
LNG FILTER CONFIRM rule
BUY rule
SHT REV FILTER CONFIRM rule
REVERSE-SELL & SELL SHORT
rule
SELL rule
What is important to understand is that you are managing here TWO exit
signals, and that you absolutely need to put your Reverse Rules Statement (which
will be activated if the associated Filter CONFIRM rule returns true) BEFORE
the usual exit rules statement… The order in which you place your trading
actions on the rules list is vital in IRT.
Please note as well that an SHT_FILTER_EOB_REV for example will be
almost identical to the SHT_FILTER_EOB one
except that we will add to it a check for “POS_STATE=1”, i.e. this
filter can only returns true in a REVERSE situation, i.e. when a LONG trade is
ongoing
Ok, this might sound a bit complex for a implementing a reverse rules,
but remind here that we are managing in fact two exit signals.
Of course, one could decrease the number of rules by incorporating the
“Filter rules” within the actual SELL and REVERSE-SELL & SELL SHORT rules.
But I really prefer this modular approach and use of CONFIRM rules.
There are further advantages that will appear only in the coming
chapters..
For the Short side, the set of rules will be:
SHT FILTER CONFIRM rule
SELL SHORT rule
LNG REV FILTER CONFIRM rule
REVERSE-COVERSHORT
& BUY rule
COVERSHORT rule
And if we combined both long and short rules:
LNG FILTER CONFIRM rule
BUY rule
SHT FILTER CONFIRM rule
SELL SHORT rule
LNG REV FILTER CONFIRM rule
REVERSE-COVERSHORT & BUY
rule
COVERSHORT rule
SHT REV FILTER CONFIRM rule
REVERSE-SELL & SELL SHORT
rule
SELL rule
You have noticed that rules for initiating new trades are of course
situated before the exit signals.
In EOB type of approach, you usually won’t have at the Close both Short
and Long trades triggering at the same time (otherwise, recheck your rules
logic!!), so the respective order between the SHT and LNG rules (within the
ENTRY or EXIT set of rules) doesn’t matter.
HOWEVER, with the INB approach, it is fully possible that the same bar
trigger a BUY and SELL actions (imagine a very wide outside bars breaking at
the same time the previous 10 bars Low and Highs). In this quite rare
situation, there is no magic, with the above set of rules, the BUY signal would
be triggere
Anyhow, as there is no way to know if the HI or LO of a bar occurre
Moreover, with our present set of rules, what would happen is that the
trade will be then systematically closed on the same bar, resulting in a
negative PL…
At least, this will not “over-optimize” our system performance…
So I consider these point to be really a very minor inconvenient
compared to the benefits of an INB approach for simulating breakouts…
Please remember as well, there is many ways to manage Filter and Reverse
rules in IRT. I am just showing you one possible approach! If you select
another approach, just be careful to have a set of coherent rules…
I have attached in 1.6.3 the full script for an EOB system including
Filter and Reverse rules
1.5) Tips
for getting EOB / INB signals markers on your chart and about the equity curve
chart
We will be discussing here quickly charting issue related to:
A) Entry and exit trade signals markers
Breakout types of entry are the easy one to program using a INB method. What is very nice with such approach is that
you do not have any blinking signals on your chart.
Here is what I mean:
Let’s say that I have a Signal marker associated to the rule:
HI > STAT.1
In this case, as soon as the marker appears once during the life of a 5
minute bars, it will remains: indeed, the High can only goes higher (except may
be in case of bad ticks, which are really rare with E-minis)…
On the other hand, let's consider the EOB rules:
CL > STAT.1
Depending on the last trade value, during the building of the 5 min
bars, you could have a “blinking marker” if the ongoing trades “oscillate”
successively above and below the trigger level. This is a good indication a
trade could be trigger soon, when the bar is completed, but is quite annoying
at the end of the day if you follow this marker to actually pull the trigger on
your broker platform.
My tip is the following in case of EOB entry signals:
1) I would use a color marker BELOW the chart for showing different colors depending on the status of the CL > STAT.1
statement
2) I use on the chart
a signal marker, but it is :
ð
associated to the signal CL.1 > STAT.2
ð
the marker is shifted one bar on the left
ð
and is linked to a given wav sound
This way, if the signal triggers indeed at the close of the bar, the
marker you then appears ONCE on the chart (will not blink of course) and a
sound alarm will tell me that I have a CONFIRMED buy signal.
There is much more to say on this charting and markers issue. Indeed,
the problem with using the embedded “Chart rules” features in IRT is that it
does not take into account the preceding CONFIRM rules. So, on my side, I will
use another approach.
I have already detailed it a bit in the Yahoo forum:
You may have a look for example at this particular message:
http://groups.yahoo.com/group/LinnSoft/message/5601
which illustrate a very interesting
use of the powerful Signal Statistics token (SSTAT)
More on that may be one day in a possible chapter 4
B) the Equity curve Chart :
OK, there is a lot to say here as well but I will try to keep it to the essential:
The equity curve information is stored in the $REF_SYSTEM_A_EOB
instrument which will be only generated if you check the box “Account Balance
charts” of the “Setup backtest” window.
If you have an intraday system (based on 5 minute bars) and if you want
to have a nice equity curve of the past 2 years (if this is your backtest
period) fitting on your screen width (with no scrolling), you need to create a
chart with
ð
a style being continuous line
ð
a low pixel per bar set up(1 or 2)
ð
a periodicity typically equal to your day session
duration in minutes (i.e. like 405 minutes)
ð
a gridline color being the
same than your background color (this is require to
make the daily start of session vertical line, which will be highly compacted
with a 405 min. periodicity) invisible
Then, I recommend to add some vertical
reference line at given date (like at every year or month start).
This way, you get a nice chart dedicate
Below the equity curve, I would suggest to chart the following
information in separate pane:
ð
The ongoing Drawdown
This is showing you at a given time the current drawdown (on a daily
basis if your periodicity is equal to the daysession duration): this curve can
be obtained using a (quite simple) custom indicator equal to “STAT – CL” with
STAT being the “Maximum Close” since a fixed date (i.e. the start date of your
backtesting period)
ð
The instrument data
Please note that the equity curve is an instrument like any other, so it
is possible to apply to it any kind of technical indicators, including custom
indicators…
The final result could look like the following screenshot
1.6) Detailed scripts of the trading systems discussed
in Chapter 1
Before going
further, here the transcript of trading system with Long and Short rules an
You will notice
that, for legibility purpose, all my Trading signals names can be cuts
in 3 part…
First part is
either __LNG / __SHT / __L_S meaning this is either a dedicated LONG / SHT
rules or a rules common to both type.
It is generally
followed by the type of action, i.e. ENTRY / Exit or FILTER
Finally, if the
trading signal is specifically linked to an EOB/INB approach or use