Hello Mike,
I introduced the forum to the concept of 'Root Cause Evaluation' (RCE)
sometime in the recent past.
Given that it is only a formative (juvenile) idea at this stage, that
I have only discussed it, in the forum, in a cursory and piecemeal
fashion and that people take time (repetition required) to
acknowledge, let alone consider, new ideas it is no surprize that
most members aren't getting it.
The basic premise is that the possible spread of future equity
outcomes, and hence metrics, can be estimated from the trade series%
and the time in trade.
The trade series%, along with it's metrics, are the profile of the
system (any system) whereas many of the things you mention (position
size/shrinking etc) are related to Portfolio Management.
Irrespective of what is done at the Portfolio Management level the
trade profile does not, and can not, change (if you use Portfolio
feedback to change your system rules then you have changed systems,
full stop).
Since there are a myriad of possible Portfolio Management (PM) rules,
IMO, it can cause a good deal of confusion by integrating System
Profile (SP) and PM in one step (which is what backtesters, buy user
demand, do).
It can help clarify trading a lot if we sometimes just focus on the
root causes.
Also, if we are going to use SP, in tandem with PM, as an inline
indicator then we might need to consider how to go about that in
efficient ways (above and beyond what is available in 'canned'
backtesters).
Example of the premise:
SystemA == predefined buy, sell, price rules etc;
Trade 1 == buy at 100, sell at 110, buy/sell == 1.1 (growthfactor),
number of bars in trade;//10% win
bars out of trade;
Trade 2 = buy at 123.71, sell at 120;//approx 3% loss
bars in trade;
etc;
We can not change a System A rule without changing the system (for
any reason) therefore the only (mathematically) possible data we can
acquire about the system is the tradeseries% and the bars in/bars out
for the trade (the values will vary from underlying to underlying but
not the available datapoints).
Since the trade rules are fixed, the trade series is also fixed for a
fixed underlying (symbol) and time period.
Logically all metrics can only be derived from these root, or core,
metrics.
Equity, or Portfolio, outcomes can vary depending on how much we
stake on the system (but the system does not, and can not change for
any given underlying series).
(Note that we can't control the underlying, only our system rules and
our staking).
Example of Equity/PM outcome:
>From the above example the trade series is 1.10.0.97 (closed trades
only).
It could have bars out of trade in between - that doesn't affect the
Eq/PM outcome, only the % return relative to time (unless we take
into consideration interest paid on idle money, margin etc which,
once again is PM).
So a hypothetical trade%series is 1,1,1.10,1,0.97,1.05 (as growth
factor).
If we consider Eq as unit 1 (standardised) the outcome is:
Eq == 1 * growth factor (1,1,1.10,1.10,1.067,1.05)//sadly I think
looping is required to produce an Eq curve???
If we use fixed capital staking (say 100 InitialEq) we have:
Eq == 100,100,110,110,106.70,112.03
Note that the relativity of the staked Eq series, to the standardised
Eq series, doesn't change, so for the purpose of comparing system
performance, on the fly (using RCE) we can do without the equity
allocation and just use the standardized array if we choose to.
If we mix and match our staking then we get different Eq/PM outcomes:
Staking == 100,no change,no change,add 100 to the stake,no change,no
change
Eq == 100,100,110,210,203.70,213.88
We end up in a much different Eq/PM place but the system itself
hasn't changed (it is no better, or worse, than it was before).
I believe that the premises of Root Cause Evaluation are sound.
What I haven't done, in this thread, is:
a) provide the code to produce the tradeseries% array
b) allow for open, or concurrent, trades
c) provide the code to record bars in/out of trade
d) prove that all possible metrics can be derived from the core
metrics
In the context of Al/Hermans thread:
- I am (they are) talking about inline performance metrics and not a
backtester (they are different things)
- they didn't ask for an inline Portfolio Manager, only inline
relative performance of a system (on the fly)
- they haven't told us 100% of what they want to do and why
(understandably)
- since they are looking for a metric indicator there is no need to
measure bars in/out of trade ( we can see that unfold, bar by bar in
the indicator). If they want that I assume it can be done but that is
personal choice
- the code needn't become too unwieldy because it would be a template
that they would apply on a metric by metric basis (presumably they
only want to watch one or two metrics at a time and not all of them)
- it is not my idea, it is theirs. I am offering them some of my
seminal ideas on RCA and it is up to them if and how they apply it
Setting aside the proofs for the argument that all metrics can be
derived from the root inputs and restricting the discussion to the
code required for the tradeseries%.
IMO the closed trade series, as % will do the job they want to do
(providing they understand the mindset of RCA).
I believe we can produce trade series arrays, given the system rules.
What I haven't done is produce an array trade series where we record
a second signal that occurs, for any symbol in our watchlist, while
we already have an open signal for that symbol but I expect that can
be done as well.
On top of that I haven't allowed for open trade values.
IMO the slight error that these 'errors' introduce into their metric
won't have a significant affect on what they are doing but if they
want to count all signals, and account for open trades they can
(there is a trade off there because it will slow down their
indicator).
As I said before, the only real 'missing link' is how to get StDev,
for a trade series, on the fly (although so far Herman and Al haven't
said they are interested in that - the idea to track trade series SD
is mine).
I assume that iff I want to get the closed trade series array I can
just go back to Freds code and use/copy/modify that, OR get something
else that is already around OR make up my own OR ask the forum for
help OR ask support for help.
I am confident it can be done, in fact I am inclined to think I could
do it myself and that is saying something.
Are you suggesting it can't?
Hope that opens up some new lines of thinking.
I appreciate I haven't done everything, that I possibly can, to
explain the concepts/practise but personal choice comes into
application anyway and there are countless variations on the
applications.
BTW
I like your work.
Cheers,
Brian_z
--- In amibroker@xxxxxxxxxps.com,
"Mike" <sfclimbers@...> wrote:
>
> Brian,
>
> I believe that the crux of the issue is the part that you only
> mentioned in passing:
>
> > "It doesn't allow for the instances where a new buy signal is
> generated on an individual equity when we are already in a trade,
OR
> scaling in or out BUT it is only at the basic conceptual stage -
> perhaps it lends itself to more sophisticated calcs if required."
>
> If I understand your post correctly, your premise hinges on the
> ability to recognize trades using the following lines:
>
> > arrayAAA = if(no sell,1,trade%); //runs on symbol AAA
> > arrayBBB = if(no sell,1,trade%);//run on symbol BBB
>
> Yet, the values to put in these arrays are the very values that
> cannot be determined in the absence of a fullblown portfolio
> backtest; One which takes into consideration such factors as
account
> balance, maximum number of positions, position scoring, position
size
> shrinking, redundent signals, scaling, custom signal cancelations
in
> CBT code, etc.
>
> This is why the distinction is being made between "doable"
individual
> symbol metrics vs. vastly more complex portfolio metrics.
>
> Mike
>
> --- In amibroker@xxxxxxxxxps.com,
"brian_z111" <brian_z111@> wrote:
> >
> > Anyone?
> >
> > Notwithstanding that Tomasz could do this stuff for us (if and
when
> > he comes back from his lovely holiday) it is fun to look at the
> > concepts involved and figure how we can do it ourselves, with
what
> we
> > have already.
> >
> > I haven't read the various code solutions closely but I have been
> > following along at a distance (all the way back to Fred's earlier
> > PerformanceMetric code).
> > Although I havent had a need for it I assumed that this stuff
could
> > be done with array programming OR maybe a loop or two (on the
fly -
> > no need to run the backtester).
> >
> > My proposition is that all metrics can be derived from the trade
> > series (as %) and the time in trade i.e. the core metrics.
> >
> > So if the trade series % is stored in an array then the array can
> be
> > used to produce the eq curve OR the metrics, on the fly, as
> required.
> >
> > (I can't remember if this is what Fred did before - it was pretty
> > close to it - anyway - I think this is a slightly different
concept
> > applied to the same problem).
> >
> >
> >
> > Simplified Example (in pidgin pseudocode):
> >
> > - this basic example only gives the closed trade metrics
> >
> > - for ease of discussion I am mentally referencing EOD data bar
by
> bar
> >
> > - two symbols in a watchlist AAA, BBB
> >
> >
> >
> > buy = whatever;
> > sell = whatever;
> >
> > sellvalue = etc;//code required
> >
> > trade% = sellvalue/buyvalue;//needs to be expressed as
growthfactor
> > i.e. 3% win == 1.03, 3%loss == 0.97 etc
> >
> > arrayAAA = if(no sell,1,trade%); //runs on symbol AAA
> > arrayBBB = if(no sell,1,trade%);//run on symbol BBB
> >
> >
> > We would end up with two arrays:
> >
> > arrayAAA == 1,1,1.10,1,1,1 etc//exited trade with 10% win on bar
3
> > arrayAAA == 1,1,1.07,1,1,1 etc//exited trade with 7% win on bar
3
> >
> >
> >
> > the above arrays can easily be used to produce other arrays with
> W/L
> > ratios etc, including for any lookback (as per Tomasz's example
b).
> >
> > To produce an eq curve (individual OR portfolio):
> >
> > combined trade% is the product of arrayAAA and arrayBBB
> >
> > arrayAAA == 1,1,1.10,1,1,1
> > arrayAAA == 1,1,1.07,1,1,1
> > combinedtrade%== 1,1,1.177,1,1,1//bar by bar product of
above two
> > arrays
> >
> > EquityCurveArray = Initial Equity * combinedtrade%;
> >
> > example:
> >
> > InitialEquity = 100;
> >
> > 100 * 1 *1 * 1.177 * 1 * 1 * 1 ==
> > EquityCurveArray == 100,100, 117.70,117.70,117.70,117.70
etc
> >
> > which is the same even if the two trades exit on separate bars.
> >
> > 100 *1.10 * 1.07 = 117.70;
> >
> > Note that the value of 1 is used to denote 'no trade' so it
doesn't
> > affect equity.
> >
> > All of the metrics required, on the fly, can be calculated using
> > lookback periods for the combinedtrade% array (it is like an
array
> > that is the carrier of the trade series) AND/OR calculating
equity
> on
> > the fly AND the code already posted OR some new relatively simple
> > array code.
> >
> > The only thing we can't do is get a standard dev of the trade
> series
> > because we can't write them into an array without all of the no
> > trades (like a static variable except a static array).
> >
> > If StDev of the trade series were could we calcuate every single
> > metric directly without the need for the eq curve?
> >
> > I think so but I haven't checked to be certain about it
(certainly
> > every worthwhile one because a lot of them are old rubbish).
> >
> > Time in trade is only needed if we want to work out percentage
> gains
> > relative to time?
> >
> > It doesn't allow for the instances where a new buy signal is
> > generated on an individual equity when we are already in a trade,
> OR
> > scaling in or out BUT it is only at the basic conceptual stage -
> > perhaps it lends itself to more sophisticated calcs if required.
> >
> >
> >
> > BTW
> >
> > I don't think AB or Tomasz is dumb - only that the trading
> community
> > in general, IMO, is bizarrely obsessive about eq curve
> > analysis/metrics.
> >
> >
> > brian_z
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > --- In amibroker@xxxxxxxxxps.com,
"Tomasz Janeczko" <groups@>
> > wrote:
> > >
> > > Hello,
> > >
> > > Yes, there are generally 3 different metrics "areas"
in terms
of
> > computational complexity:
> > >
> > > a) metrics derived directly from Equity (CAR, Drawdowns, UPI,
> > Sharpe, Std error, etc)
> > > - they can be easily calculated directly from Equity data
> > (~~~EQUITY) as shown in the
> > > code that Fred posted below.
> > > There is no computational difference between calculating them
on
> > portfolio or individual
> > > symbol basis.
> > >
> > > b) metrics NOT derived from equity (number of winners,
> winner/loser
> > ratio, expectancy, etc)
> > > - these metrics needs actual backtest to be performed to get
the
> > TRADE LIST and
> > > perform calculations based on trade list.
> > > Performing backtest on SINGLE security is done by Equity()
> function
> > and
> > > single-security metric "indicators" are possible with
the same
> > > computational complexity as Equity() function has.
> With "possible"
> > I mean that I could implement that
> > > if there is enough demand.
> > >
> > > c) same as b) BUT for Portfolio-level backtest. Again you need
a
> > trade list to calculate them.
> > > No matter which approach is used, full-blown portfolio backtest
> for
> > specified "lookback period"
> > > is required. It is too complex to be done on-the-fly in real
> time.
> > Hence proposed solution of using
> > > CBI to generate stats and then refer to them using Foreign()
> > function.
> > >
> > >
> > > Best regards,
> > > Tomasz Janeczko
> > > amibroker.com
> > > ----- Original Message -----
> > > From: Fred Tonetti
> > > To: amibroker@xxxxxxxxxps.com
> > > Sent: Tuesday, July 29, 2008 6:16 AM
> > > Subject: RE: [amibroker] Re: How to save Metrics in
Composites
> > for Individual BTs
> > >
> > >
> > > . and if you are wanting the performance metrics as arrays
over
> > some shorter length then the backtest period and you aren't
wanting
> > to do it in the CBI the while you're waiting for it to be
available
> > as a built in AFL function you can fill in the rest of this .
Both
> TJ
> > in an earlier post and the DOC have the formulas for the rest of
> the
> > performance metrics .
> > >
> > > MACDx = MACD(12, 26);
> > >
> > > MACDsig = Signal(12, 26, 9);
> > >
> > >
> > >
> > > Plot (MACDx, "MACD", colorRed);
> > >
> > > Plot (MACDsig, "Sig", colorWhite);
> > >
> > >
> > >
> > > Buy = Cross(MACDx, MACDsig);
> > >
> > > Sell = Cross(MACDsig, MACDx);
> > >
> > >
> > >
> > > function getMetric(Metric, Length)
> > >
> > > {
> > >
> > > Metric = StrToUpper(Metric);
> > >
> > > BIR = IIf(Status("BarInRange") > 0, 1,
Null);
> > >
> > > CurEq = Foreign("~~~EQUITY", "C", 1);
> > >
> > > MaxEq = HHV(CurEq, Length);
> > >
> > > CurDD = IIf(CurEq < MaxEq, 100 * (MaxEq - CurEq) /
> MaxEq,
> > 0);
> > >
> > > MaxDD = HHV(CurDD, Length);
> > >
> > > SqrDD = CurDD ^ 2;
> > >
> > > SumDD = Sum(SqrDD, Length);
> > >
> > > UI = sqrt(SumDD / Length);
> > >
> > > if (Metric == "CDD")
> > >
> > > Value = CurDD;
> > >
> > > else if (Metric == "MDD")
> > >
> > > Value = MaxDD;
> > >
> > > else if (Metric == "UI")
> > >
> > > Value = UI;
> > >
> > > else
> > >
> > > Value = Null;
> > >
> > > return Value;
> > >
> > > }
> > >
> > >
> > >
> > > Plot(getMetric("UI", 20), "UI",
colorYellow,
> styleLeftAxisScale);
> > >
> > >
> > >
> > >
> > >
> > >
> > > ----------------------------------------------------------
--
> --
> > ----------
> > >
> > > From: amibroker@xxxxxxxxxps.com
> > [mailto:amibroker@xxxxxxxxxps.com]
On Behalf Of Paul Ho
> > > Sent: Monday, July 28, 2008 11:59 PM
> > > To: amibroker@xxxxxxxxxps.com
> > > Subject: RE: [amibroker] Re: How to save Metrics in
Composites
> > for Individual BTs
> > >
> > >
> > >
> > > Herman
> > > I really can't see the difficulty to achieve any of these,
see
> my
> > reply
> > > below.
> > >
> > > ________________________________
> > >
> > > From: amibroker@xxxxxxxxxps.com
> [mailto:amibroker@xxxxxxxxxps.com]
> > > On Behalf Of Herman
> > > Sent: Tuesday, 29 July 2008 11:32 AM
> > > To: amibroker@xxxxxxxxxps.com
> > > Subject: RE: [amibroker] Re: How to save Metrics in
Composites
> for
> > > Individual BTs
> > >
> > >
> > >
> > > Hello Fred,
> > >
> > > Putting technical problems and limitations aside it is clear
to
> me
> > > that having access to period-based performance metrics is
> > critical in
> > > trading.
> > >
> > > [I don't see a lot of technical difficulty. See my previous
> post
> > on ONE way
> > > to do this. There probably a few other ways to do it I can
> think
> > of.]
> > >
> > > Since the Equity curve is based on system performance it
makes
> > common sense
> > > that performance metrics will lead on the Equity. Most
> > performance metrics
> > > can't be extracted from the equity curve; they are lost at
that
> > > summary-stage. For example, DrawDown is the result of
advanced
> > system
> > > failure; it might very well have been possible to detect the
> > upcoming
> > > failure (DD) by looking at some very basic metrics. Simply
> > looking at the
> > > number of winning trades or profit/trade, might have alerted
> us.
> > Using more
> > > sophisticated metrics, like a period based UPI Indicator, we
> > might have seen
> > > a very clear and gradual system failure. All this is
especialy
> > true for
> > > faster trading systems where metrics become very smooth
> > indicators.
> > >
> > > To look at metrics for the entire duration of the backtest
may
> > come
> > > in handy during development however during real trading it is
> > useless. It is
> > > simply due to a reluctance to break with tradition (a problem
> for
> > most
> > > traders) that we don't have period-based performance
indicators
> > today. They
> > > are long overdue. Sure, they they are 'doable' in AmiBroker,
> but
> > how many
> > > users would be able to write such code? In my opinion not
more
> > that 1%. To
> > > develop systems efficiently we need basic afl tools; no one
> wants
> > to spend
> > > two months learning OLE only to find out that it doesn't fill
> the
> > > requirement or is to complicated for the user. Same for the
> CBT,
> > who wants
> > > to spend two months learning it just to test and idea that
may
> or
> > may not
> > > work? Time is precious...we want to trade, not become
> professional
> > > programmers.
> > >
> > > [firstly it doesn't take 2 months, probably wont even be 2
> weeks.
> > Secondly
> > > it doesn't have to involve OLE as far as I can gather, though
> OLE
> > is clearly
> > > useful in other area that I am finding out for myself.
Thirdly,
> > trading is a
> > > lifetime persue, not a five minute flash in the pan. While
only
> > Tomasz will
> > > know fully, I suspect, do what you say as a "simple"
function
> > outside CBT
> > > would involve storing heaps of unnecessary data, Who is then
> > paying for the
> > > penalty of that?. Inside CBT its relatively simple to create.
> But
> > you will
> > > still need to learn CBT. By the way, Is there any other
> software
> > out there
> > > that can do this sort of things. And in AB it is readily
> > available, with
> > > just a few lines. What is the problem?]
> > >
> > >
> > > A few common applications for metric indicators are:
> > >
> > > 1) When trading a number of different systems one wants to
know
> > > early when a system starts to fail. How else can one make a
> > timely switch?
> > > Just like you may want to trade funds with a rotational
system
> > you want to
> > > rotate the trading systems themselves.
> > > 2) It is very common for trading systems to fade in and out of
> > > performance. To detect when a trading system fails and to
> switch
> > to another
> > > system requires, again, monitoring system performance
metrics.
> > imo, There is
> > > just no other way around it.
> > > 3) And of course we want to know when a ticker loses its
> character
> > > and stops working. Here again performance metrics might be
the
> > best way to
> > > detect ticker failure.
> > >
> > > A function like getPerformanceMetric ( MetricName,
> LookBackPeriod)
> > > would find wide application. Most people think right away of
> > portfolio
> > > trading however performance metrics should be calculated for
> > individual
> > > stocks, they will lose their meaning if derived from
portfolio
> > results.
> > >
> > > I am not saying that it is easy to design such indicators,
> perhaps
> > > it is extremely difficult. But that doesn't remove the need
for
> > them.
> > >
> > > Just my two cents worth :-)
> > > Best regards,
> > > herman
> > >
> > > [a lot of what you have asked for can be readily derived from
> the
> > equity
> > > curve, without even going through CBT, while the rest is a
just
> a
> > few lines
> > > in CBT, plus a function or two outside. Recently I helped
> > somebody code the
> > > UI for individual stocks, 5 lines of AFL, that's all. I just
> > wonder if this
> > > is a mountain, or a mole hill]
> > >
> > > -----Original Message-----
> > > From: amibroker@xxxxxxxxxps.com
> > > [mailto:amibroker@xxxxxxxxxps.com]On
Behalf Of Fred Tonetti
> > > Sent: July 28, 2008 12:44 PM
> > > To: amibroker@xxxxxxxxxps.com
> > > Subject: [SPAM]RE: [amibroker] Re: How to save Metrics in
> > > Composites for Individual BTs
> > >
> > >
> > >
> > > Al,
> > >
> > > What you are looking for in AB is I believe
> > > a little more difficult then or at least as time consuming as
> you
> > think it
> > > is .
> > >
> > > Share ? . Code ?
> > >
> > > LOL . I thought I did in that post .
> > >
> > > Ok . Kidding aside . Assuming that
> > >
> > > - AB is running
> > >
> > > - AA has the AFL you want to run ( This doesn't
> > > mean the formula editor as the FE and AA aren't coupled )
> > >
> > > - AA Settings for ApplyTo / Range etc are as you
> > > want then
> > >
> > > This VBS code would run a BackTest and
> > > export the Trade List to a File
> > >
> > > Dim oAB
> > >
> > > Dim oAA
> > >
> > > Set oAB = CreateObject("Broker.Application")
> > >
> > > Set oAA = oAB.Analysis
> > >
> > > oAA.Backtest(0)
> > >
> > > oAA.Export("Dummy.csv")
> > >
> > > The code above should be saved in a .vbs
> > > filetype ( name of your choosing ) and then simply double
> > clicking it will
> > > produce the file.
> > >
> > > This could easily be changed to running an
> > > optimize which will produce all the performance metrics in
the
> AA
> > results
> > > which could then be exported by changing .
> > >
> > > oAA.Backtest(0)
> > >
> > > to .
> > >
> > > oAA.Optimize(0)
> > >
> > > Ideally the optimize above would be a one
> > > step optimize if you will just to produce the performance
> metrics
> > related to
> > > the backtest .
> > >
> > > With a little more work i.e. a loop to set
> > > the beginning and ending dates one could get performance
> metrics
> > externally
> > > for whatever lookback length one wanted one after the other .
> > >
> > >
> > > ________________________________
> > >
> > > From: amibroker@xxxxxxxxxps.com
> > > [mailto:amibroker@xxxxxxxxxps.com]
On Behalf Of Al Venosa`
> > > Sent: Monday, July 28, 2008 12:17 PM
> > > To: amibroker@xxxxxxxxxps.com
> > > Subject: Re: [amibroker] Re: How to save Metrics in
> > > Composites for Individual BTs
> > >
> > > Dear TJ and Fred:
> > >
> > > Thank you for these suggestions. However, what I am asking
> > > for is a simple AFL function in which we can specify the
> lookback
> > period for
> > > the metric in question, nothing more. To observe the
> difference,
> > you can
> > > simply substitute an MA() for the required function, as shown
> > below. When
> > > you view the graph, you will see, I think, what I am talking
> > about.
> > >
> > > Plot(C,"",1,128);
> > > Plot(MA(C,50),"FixedLookBack",colorRed,1);
// the requested
> > > solution
> > > Plot(MA(C,BarIndex()),"ExistingNow",colorBlue,1);
// TJ's
> > > and Fred's solution.
> > >
> > > I believe what TJ has suggested, if I am interpreting it
> > > correctly, is that I would be running a backtest for each bar
> in
> > a loop,
> > > which would be complex and very slow. Maybe I'm asking the
same
> > thing. If
> > > so, tell me, and I'll desist. An alternative solution is to
use
> > the Walk
> > > Forward Individual Backtester to implement a system that uses
> > performance
> > > metrics as position score.
> > >
> > > BTW, Fred, are you willing to share the complete working
> > > code for your 2-line export?
> > >
> > > Thank you.
> > >
> > > Al V.
> > >
> > > On 7/28/08, Tomasz Janeczko
> > > <groups@ <mailto:groups@> > wrote:
> > >
> > > Hello,
> > >
> > > It is doable with custom backtester and not
> > > so complicated.
> > >
> > > As described in detail here:
> > >
> > > http://www.amibroker.com/guide/a_custommetrics.html
> > > <http://www.amibroker.com/guide/a_custommetrics.html>
> > >
> > > You have direct access to ANY backtest
> > > performance metric using
> > >
> > > GetPerformanceStats() function of backtester object.
> > >
> > > There is no obstacle in calling it every bar
> > > and storing the result in the array if you wish.
> > >
> > > Let say you want UPI as array.
> > >
> > > // your trading system here
> > > Buy = ...
> > >
> > > Sell = ...
> > >
> > > SetCustomBacktestProc("");
> > >
> > > /* Now custom-backtest procedure follows */
> > >
> > > if( Status("action") == actionPortfolio )
> > > {
> > > UPI = 0;
> > > bo = GetBacktesterObject();
> > >
> > > bo.PreProcess();
> > >
> > > for( bar = 0; bar < BarCount; bar++ )
> > > {
> > > bo.ProcessTradeSignals( bar );
> > >
> > > st = bo.GetPerformanceStats(0); // get stats for all
> > > trades
> > >
> > > UPI[ bar ] = st.GetValue("UlcerPerformanceIndex");
> > > }
> > >
> > > bo.PostProcess();
> > >
> > > AddToComposite( UPI, "~~~UPI", "X",
atcFlagDefaults |
> > > atcFlagEnableInPortfolio );
> > > }
> > >
> > > Now ~~~UPI ticker will contain bar-by-bar
> > > values of Ulcer Performance Index.
> > >
> > > As for "specifying lookback period" it is
> > > doable by creating Xth composites (and X backtests) each
> > containing values
> > > for specified
> > >
> > > lookback period.
> > >
> > > As for Equity() - this is SINGLE security
> > > (OLD) backtest. It has no comparision to portfolio level
> backtest
> > that
> > >
> > > must go through entire portfolio. The complexity of
> > > portfolio backtest is Nth times the single security backtest
> > >
> > > where N is number of symbols in portfolio. Therefore it is
> > > not feasible to be calculated on-the-fly in real time like
> > >
> > > single-security backtest (i.e. Equity()).
> > >
> > >
> > > Best regards,
> > > Tomasz Janeczko
> > > amibroker.com
> > >
> > > ----- Original Message -----
> > >
> > > From: Al Venosa` <mailto:avcinci@>
> > >
> > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker@xxxxxxxxxps.com>
> > >
> > > Sent: Monday, July 28, 2008
> > > 1:02 AM
> > >
> > > Subject: Re: [amibroker] Re: How to save Metrics in
> > > Composites for Individual BTs
> > >
> > > Thanks, Fred, but I haven't
> > > a clue how to do OLE/Automation, and I wonder how many AB
users
> > out there
> > > really do. That's why I was calling for a simple, non-
painful,
> > easy-to-use
> > > AFL function that would do this for the non-
techie/programmer,
> > and if it
> > > existed, AB would be the only trading software out there that
> > would be able
> > > to do this. I bet It would be a profit bonanza for TJ.
> > >
> > > Al V.
> > >
> > > On 7/27/08, Fred Tonetti
> > > <ftonetti@ <mailto:ftonetti@> > wrote:
> > >
> > > I agree that this would be nice to have as directly
> > > as you have laid out .
> > >
> > > However, while somewhat
> > > painful, one can with Equity() and a list of trades calculate
> all
> > the
> > > performance metrics as of any given bar or if you prefer as
> > arrays of values
> > > .
> > >
> > > This could be fully
> > > automated with OLE/Automation
> > >
> > >
> > > ________________________________
> > >
> > > From:
> > > amibroker@xxxxxxxxxps.com
<mailto:amibroker@xxxxxxxxxps.com>
> > > [mailto:amibroker@xxxxxxxxxps.com
> > <mailto:amibroker@xxxxxxxxxps.com>
] On
> > > Behalf Of Al Venosa
> > > Sent: Sunday, July 27, 2008 11:29 AM
> > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker@xxxxxxxxxps.com>
> > > Subject: [amibroker] Re: How to save Metrics in
> > > Composites for Individual BTs
> > >
> > > Having read this interesting
> > > thread begun last week, I'd like to
> > > continue it with a follow-on question/comment that I
> > > think, if TJ were
> > > to implement it, would make Amibroker infinitely
> > > more useful to the
> > > actual trader. It would be awesome if we could have
> > > simple-to-use AFL
> > > functions that read AFL backtest metrics directly.
> > > Adding a lookback
> > > period would make them immensely more useful as
> > > indicators. What I am
> > > suggesting is to have a function like:
> > >
> > > getEquityMetric ( MetricName,LookBackPeriod);
> > >
> > > The example code provided by TJ gives us only the
> > > cumulative value of
> > > each metric. What I'm suggesting is to go beyond
> > > this one cumulative
> > > output number and create metric arrays with a
> > > specified lookback
> > > period. Then, when we plot these metrics in an
> > > indicator, we can
> > > visually look for correlations with price charts.
> > > For example, one
> > > could plot winning trades/month and see if they
> > > change with trend. Or
> > > one could look at AverageWin or UPI and see how that
> > > changes with
> > > trend. These are all correlations that are best
> > > analyzed visually (in
> > > an Indicator) but can ONLY be analyzed if we have
> > > access to these
> > > metrics for variable lookback periods.
> > >
> > > Another use for these functions would be as a
> > > positionscore in a
> > > trading system. What better way is there to select
> > > tickers/systems to
> > > trade than the actual performance of those tickers
> > > or systems? The
> > > procedure may require a preliminary scan/exploration
> > > to create metric-
> > > composites that can be read by the trading system
> > > and used as a
> > > positionscore. Critical here is that the metric can
> > > be read for any
> > > specified lookback period, i.e. 10 bars, 100 bars,
> > > etc. So the
> > > function must have a period argument, which is the
> > > most important
> > > factor. We already have equity(). Why not expand
> > > this with the other
> > > backtest metrics?
> > >
> > > Undoubtedly, all this can be implemented using the
> > > custom backtester,
> > > but this solution probably excludes >95% of all
> > > AmiBroker users.
> > >
> > > TJ, would this be possible to implement?
> > >
> > > Al Venosa
> > >
> > > --- In amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com> ,
"Herman" <psytek@>
wrote:
> > > >
> > > > Thank you TJ :-) you saved the day once more !
> > > >
> > > > Great stuff.
> > > >
> > > > If someone is wondering why I wanted this
> > > program... You can design
> > > trading
> > > > systems and use performance metric arrays as
> > > powerful Indicators. It
> > > is
> > > > somewhat similar to trading the equity curve.
> > > Price arrays can have
> > > > qualities that can make your trading systems fail
> > > but that are
> > > undetectable
> > > > with traditional indicators.
> > > >
> > > > However, you can design small trading systems that
> > > target specific
> > > price
> > > > characteristics, like patterns, trends,
> > > volatility, cycles, etc.
> > > Using the
> > > > code below gives you statistical information about
> > > these
> > > characteristics in
> > > > a form that can be plotted, and be used in other
> > > trading systems.
> > > >
> > > > Thanks everyone for your help!
> > > > have a great trading day!
> > > > herman
> > > >
> > > > // Demo trading system
> > > > Short = Cover = 0;
> > > > Buy = Cross( MACD(), Signal() );
> > > > Sell = Cross( Signal(), MACD() );
> > > > // Using the CBT to retrieve/save metrics
> > > > if( Status("action") == actionBacktest )
> > > StaticVarSetText( "Symbol",
> > > > Name() );
> > > > SetOption( "UseCustomBacktestProc", True );
> > > > if ( Status( "action" ) == actionPortfolio )
> > > > {
> > > > bo = GetBacktesterObject();
> > > > bo.PreProcess();
> > > > MyHistStat1 = Null;
> > > > for ( bar = 0; bar < BarCount; bar++ )
> > > > {
> > > > bo.ProcessTradeSignals( bar );
> > > > stats = bo.GetPerformanceStats( 0 );
> > > > MyHistStat1[ bar ] = stats.GetValue( "UlcerIndex"
> > > ); // any metric
> > > can be
> > > > retrieved
> > > > }
> > > > bo.PostProcess();
> > > > AddToComposite( MyHistStat1, "~~~UI_" +
> > > StaticVarGetText
> > > ( "Symbol" ), "X",
> > > > atcFlagEnableInPortfolio | atcFlagDefaults );
> > > > }
> > > > PlotForeign( "~~~UI_"+Name(),
"UlcerIndex
> > > Historical", colorRed,
> > > > styleLine );
> > > >
> > > > -----Original Message-----
> > > > From: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > [mailto:amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com> ]On
> > > Behalf
> > > > Of Tomasz Janeczko
> > > > Sent: July 25, 2008 5:49 AM
> > > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > > > Subject: [SPAM]Re: [SPAM]Re: [amibroker] How to
> > > save Metrics in
> > > Composites
> > > > for Individual BTs
> > > >
> > > >
> > > > Herman,
> > > >
> > > > You forgot the CORRECTION I mentioned:
> > > >
> > > > StaticVarSetText( "Symbol", Name() ); must NOT be
> > > called
> > > unconditionally,
> > > > but THIS way:
> > > >
> > > >
> > >
> >
>
=====================================================================
> > > > if( Status("action") == actionBacktest )
> > > StaticVarSetText( "Symbol",
> > > > Name() );
> > > >
> > > ==============================================================
> > > >
> > > > Best regards,
> > > > Tomasz Janeczko
> > > > amibroker.com
> > > > ----- Original Message -----
> > > > From: Herman
> > > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > > > Sent: Friday, July 25, 2008 11:36 AM
> > > > Subject: RE: [SPAM]Re: [amibroker] How to save
> > > Metrics in Composites
> > > for
> > > > Individual BTs
> > > >
> > > >
> > > > Still NO GO.
> > > > I am loading the code in the AA, select a
> > > watchlist, run an
> > > Individual
> > > > backtest, and Refresh the WorkSpace. I get the BT
> > > report with
> > > individual
> > > > results. I get two Composites in my Composites
> > > Group. One is named
> > > > ~~~EQUITY, the other ~~~UI_~~~EQUITY. The first
> > > makes sense but the
> > > second
> > > > indicates that the StaticVar does not return the
> > > ticker name.
> > > >
> > > > >> It appears that in this code the function
> > > Name() returns
> > > "~~~EQUITY" and
> > > > does not return the name for the ticker being
> > > tested, it behaves as
> > > if the
> > > > ~~~EQUITY composite is the ticker being tested.
> > > > Can anyone confirm this?
> > > >
> > > > Thanks again!
> > > > Herman
> > > >
> > > > // Demo trading system
> > > > Short = Cover = 0;
> > > > Buy = Cross( MACD(), Signal() );
> > > > Sell = Cross( Signal(), MACD() );
> > > > // Using the CBT to retrieve/save metrics
> > > > StaticVarSetText( "Symbol", Name() );
> > > > SetOption( "UseCustomBacktestProc", True );
> > > > if ( Status( "action" ) == actionPortfolio )
> > > > {
> > > > bo = GetBacktesterObject();
> > > > bo.PreProcess();
> > > > MyHistStat1 = Null;
> > > > for ( bar = 0; bar < BarCount; bar++ )
> > > > {
> > > > bo.ProcessTradeSignals( bar );
> > > > stats = bo.GetPerformanceStats( 0 );
> > > > MyHistStat1[ bar ] = stats.GetValue( "UlcerIndex"
> > > ); // any metric
> > > can be
> > > > retrieved
> > > > }
> > > > bo.PostProcess();
> > > > AddToComposite( MyHistStat1, "~~~UI_" +
> > > StaticVarGetText
> > > ( "Symbol" ), "X",
> > > > atcFlagEnableInPortfolio | atcFlagDefaults );
> > > > }
> > > > PlotForeign( "~~~UI_"+Name(),
"UlcerIndex
> > > Historical", colorRed,
> > > > styleLine );
> > > >
> > > >
> > > >
> > > >
> > > > -----Original Message-----
> > > > From: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > [mailto:amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com> ]On
> > > Behalf
> > > > Of Tomasz Janeczko
> > > > Sent: July 25, 2008 4:08 AM
> > > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > > > Subject: [SPAM]Re: [amibroker] How to save Metrics
> > > in Composites for
> > > > Individual BTs
> > > >
> > > >
> > > > It will work OK.
> > > > Individual backtest *is* portfolio backtest but
> > > just portfolio
> > > consisting of
> > > > one symbol at a time.
> > > >
> > > > Note that one should select "Individual Backtest"
> > > (not "OLD"
> > > backtest) from
> > > > AA->Backtest drop down.
> > > >
> > > > One correction though
> > > > StaticVarSetText( "Symbol", Name() );
> > > >
> > > > should be called only when NOT in portfolio mode
> > > >
> > > > so
> > > >
> > > > if( Status("action") == actionBacktest )
> > > StaticVarSetText( "Symbol",
> > > > Name() );
> > > >
> > > > // Demo trading system
> > > > Short = Cover = 0;
> > > > Buy = Cross( MACD(), Signal() );
> > > > Sell = Cross( Signal(), MACD() );
> > > >
> > > > // Using the CBT to retrieve/save metrics
> > > > SetOption( "UseCustomBacktestProc", True );
> > > > if ( Status( "action" ) == actionPortfolio )
> > > > {
> > > > bo = GetBacktesterObject();
> > > > bo.PreProcess();
> > > > MyHistStat1 = Null;
> > > > for ( bar = 0; bar < BarCount; bar++ )
> > > > {
> > > > bo.ProcessTradeSignals( bar );
> > > > stats = bo.GetPerformanceStats( 0 );
> > > > MyHistStat1[ bar ] = stats.GetValue( "UlcerIndex"
> > > ); // any metric
> > > can be
> > > > retrieved
> > > > }
> > > > bo.PostProcess();
> > > > AddToComposite( MyHistStat1, "~~~UI_" +
> > > StaticVarGetText( "Symbol" )
> > > +
> > > > "_HISTORICAL", "X",
atcFlagEnableInPortfolio |
> > > atcFlagDefaults );
> > > > }
> > > >
> > > > ----- Original Message -----
> > > > From: Paul Ho
> > > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > > > Sent: Friday, July 25, 2008 4:45 AM
> > > > Subject: RE: [amibroker] How to save Metrics in
> > > Composites for
> > > Individual
> > > > BTs
> > > >
> > > >
> > > > First of all. You use Status{"action") ==
> > > actionPortfolio,
> > > individual
> > > > backtest wont go through there.
> > > >
> > > >
> > > >
> > > > From: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > [mailto:amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com> ]
> > > On Behalf
> > > > Of Herman
> > > > Sent: Friday, 25 July 2008 9:31 AM
> > > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > > > Subject: Re: [amibroker] How to save Metrics in
> > > Composites for
> > > Individual
> > > > BTs
> > > >
> > > >
> > > > Thank you Tomasz, but this code still does not
> > > work. I changed the
> > > StaticVar
> > > > to the Text type.
> > > >
> > > > Can you help some more ... ? or does anyone else
> > > see the problem?
> > > >
> > > > TIA,
> > > > Herman
> > > >
> > > > StaticVarSetText( "Symbol", Name() );
> > > > // Demo trading system
> > > > Short = Cover = 0;
> > > > Buy = Cross( MACD(), Signal() );
> > > > Sell = Cross( Signal(), MACD() );
> > > >
> > > > // Using the CBT to retrieve/save metrics
> > > > SetOption( "UseCustomBacktestProc", True );
> > > > if ( Status( "action" ) == actionPortfolio )
> > > > {
> > > > bo = GetBacktesterObject();
> > > > bo.PreProcess();
> > > > MyHistStat1 = Null;
> > > > for ( bar = 0; bar < BarCount; bar++ )
> > > > {
> > > > bo.ProcessTradeSignals( bar );
> > > > stats = bo.GetPerformanceStats( 0 );
> > > > MyHistStat1[ bar ] = stats.GetValue( "UlcerIndex"
> > > ); // any metric
> > > can be
> > > > retrieved
> > > > }
> > > > bo.PostProcess();
> > > > AddToComposite( MyHistStat1, "~~~UI_" +
> > > StaticVarGetText( "Symbol" )
> > > +
> > > > "_HISTORICAL", "X",
atcFlagEnableInPortfolio |
> > > atcFlagDefaults );
> > > > }
> > > > PlotForeign( "~~~UI_HISTORICAL",
"UlcerIndex
> > > Historical", colorRed,
> > > > styleLine );
> > > >
> > > > -----Original Message-----
> > > > From: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > [mailto:amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com> ]On
> > > Behalf
> > > > Of Tomasz Janeczko
> > > > Sent: July 24, 2008 3:00 PM
> > > > To: amibroker@xxxxxxxxxps.com
> > > <mailto:amibroker%40yahoogroups.com>
> > > > Subject: [SPAM]Re: [amibroker] How to save Metrics
> > > in Composites for
> > > > Individual BTs
> > > >
> > > >
> > > > The same code. The only distinction is that you
> > > need to run
> > > INDIVIDUAL
> > > > backtest
> > > > and use Static variable to save name
> > > >
> > > > StaticVarSet Text ("Symbol", Name() );
> > > > // Demo trading system
> > > > Short = Cover = 0;
> > > > Buy=Cross( MACD(), Signal() );
> > > > Sell=Cross( Signal(), MACD() );
> > > >
> > > > // Using the CBT to retrieve/save metrics
> > > > SetOption("UseCustomBacktestProc", True
);
> > > > if( Status("action") == actionPortfolio )
> > > > {
> > > > bo = GetBacktesterObject();
> > > > bo.PreProcess();
> > > > MyHistStat1 = Null;
> > > >
> > > > for(bar=0; bar < BarCount; bar++)
> > > > {
> > > > bo.ProcessTradeSignals( bar );
> > > > stats = bo.GetPerformanceStats( 0 );
> > > > MyHistStat1[ bar ] = stats.GetValue("UlcerIndex");
> > > // any metric
> > > can be
> > > > retrieved
> > > > }
> > > >
> > > > bo.PostProcess();
> > > > AddToComposite( MyHistStat1, "~~~UI_" +
> > > StaticVarGet Text
> > > ("Symbol") +
> > > > "_HISTORICAL", "X",
atcFlagEnableInPortfolio |
> > > atcFlagDefaults );
> > > > }
> > > >
> > > > PlotForeign("~~~UI_HISTORICAL",
"UlcerIndex
> > > Historical", colorRed,
> > > > styleLine );
> > > >
> > >
> > >
> > > ________________________________
> > >
> > > I am using the free version
> > > of SPAMfighter for private users.
> > > It has removed 512 spam emails to date.
> > > Paying users do not have this message in their
> > > emails.
> > > Try SPAMfighter <http://www.spamfighter.com/len>
> > > for free now!
> > >
> > >
> > >
> > >
> > >
> > >
> > > ________________________________
> > >
> > > I am using the free version of SPAMfighter for private
> > > users.
> > > It has removed 512 spam emails to date.
> > > Paying users do not have this message in their emails.
> > > Try SPAMfighter <http://www.spamfighter.com/len>
for free
> > > now!
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > ----------------------------------------------------------
--
> --
> > ----------
> > > I am using the free version of SPAMfighter for private users.
> > > It has removed 513 spam emails to date.
> > > Paying users do not have this message in their emails.
> > > Try SPAMfighter for free now!
> > >
> >
>