Econometrics: let's discuss the CU balance sheet.

New comment
 

I should point out straight away that I don't understand the tester results very well, so I use a different balance specification. I suggest to discuss.

So, let's take EURUSD H1 from 19.03.2012 to 28.04.2012. Here is the chart:

A certain trading system, making 47 trades, got the results (horizontal squares are out of the market):

1. As we can see, the result is estimated not in the account currency, but in pips. This is the first difference from the tester and allows you to disassociate from the size of the lot and the size of the deposit.

The result of trading for the month is 369 pips. Clearly not bad for 1.5 months of trading. But is it really that good.

Profit factor = 1.34. This figure personally makes me tense. Let's look at this figure from the other side.

2. Sum all upward movements (profit) and all downward movements (loss). We obtain the figures: up movement = 2558 pips, while down movement = 1907 pips. It turns out that the first figure is the upper limit of profitability of my TS = 2558 pips. Wow. But the second figure is the loss potential of my TS?! Also amazing.

3. Now my not bad profit of 369 pips is clearly lost amongst the potency, both positive and negative, that my TS has.

Does pp.2 and 3 tell me anything?

But that's it.

Potential of profit and loss are theoretical maximums. There is such a notion as "drawdown". Let's calculate the analogue.

The balance represents a random value. As a standard, let's exclude the deterministic component in order to estimate the random component. I smooth it out using Hodrick-Prescott filter. I get a graph.

Below is noise that shows the deviations from the smoothed value - no theoretical gains and losses.

4. The last step. Evaluate deviations. Calculate statistics. We obtain the result:

From these statistics follows:

TP = 81 pips

SL = 88 pips

Trade decisions can be made if there is a deviation from the smoothing by 14 pips

Very interesting view of the distribution. It has a huge kurtosis. Let's build a theoretical distribution:

We can see that the noise is "squeezed" to its average. This is certainly a very valuable property of TC. No surprises should be expected from it and the modest profit factor looks more attractive.

.

For the judgement of the forum. What do we think about this TS analysis? Can we use this TS?

 

IMHO, scrap the whole thing

if the MOU is greater than the "risk-free" market rate - use

 
Demi:

IMHO, scrap the whole thing

if the TC MO is greater than the "risk-free" market rate - use


I don't get it.

What is the MO TS?

There is a profit factor = 1.34

 

a profit factor of 1.34 is much more than a "risk-free" bet

for home use - use

for a level such as banks, funds, etc. - you have to adjust the profit factor for the level of risk(Standard Deviation)

 
Did you subtract the spread when you calculated in pips?
 
faa1947, what you are doing is some kind of assessment of the "goodness" of the system, and this is a relative matter. I.e. you have to compare it with other systems. But first of all you need to make sure that the system is robust. I.e. the results are not random and therefore there is hope that the profit potential (edge) will be maintained for some time. In this respect PF is not very indicative - it may be high on random trades especially if there are not too many of them. 47 trades is nothing at all for such stats. There are estimation methods that work for such a small number of trades.
 
Avals:
faa1947, what you are doing is some kind of assessment of the "goodness" of the system, and this is a relative matter. I.e. you have to compare it with other systems. But first of all you need to make sure that the system is robust. I.e. the results are not random and therefore there is hope that the profit potential (edge) will be maintained for some time. In this respect PF is not very indicative - it may be high on random trades especially if there are not too many of them. 47 trades is nothing at all for such stats. There are estimation methods that work for such a small number of trades.

which methods?
 
Demi:

what kind of methods?

secret))
 
ZZZEROXXX:
Did you subtract the spread when you calculated in pips?
No. But neither did the MM. Such a "sterile" estimate.
 
Avals:

secret))

why write meaningless posts?
 
Avals:
faa1947, what you are doing is some kind of assessment of the "goodness" of the system, and this is a relative matter. I.e. you have to compare it with other systems. But first of all you need to make sure the system is robust. I.e. the results are not random and therefore there is hope that the profit potential (edge) will be maintained for some time. In this respect PF is not very indicative - it may be high on random trades especially if there are not too many of them. 47 trades is nothing at all for such stats. There are estimation methods that work for such a small number of trades.

I think the robustness of the system is largely determined by the behaviour of the balance line error. here is the result of the stationarity test for the error.

Null hypothesis: The random component is not stationary

Exogenous: Constant

Bandwidth: 159 (Newey-West automatic) using Bartlett kernel

................................... Adj. t-Stat ...........Prob.

* Phillips-Perron test statistic -30.98050 .........0.0000

I.e. we can strictly reject the hypothesis of non-stationarity of the balance line error. It follows that we can expect no larger slippage than indicated above.



123456789101112131415161718192021222324252627282930
New comment
Reason: