Did some more investigation about it and found out that indeed it is a bug - resampler just don't work at all for timeframe == days. To fix:
The code snippet I placed in previous message:

 self.componly = (self.subdays and
                         data._timeframe == self.p.timeframe and
                         not (self.p.compression % data._compression)
                         )
...
self.componly = self.componly and not self.subdays

should be changed to (add 'not' before self.subdays and remove the last line):

self.componly = (not self.subdays and
                         data._timeframe == self.p.timeframe and
                         not (self.p.compression % data._compression)
                         )

Additionally in resamplefilter.py::Resampler::call() change this:

        if consumed:
            self.bar.bupdate(data)  # update new or existing bar
            data.backwards()  # remove used bar

        # if self.bar.isopen and (onedge or (docheckover and checkbarover))
        cond = self.bar.isopen()
        if cond:  # original is and, the 2nd term must also be true
            if not onedge:  # onedge true is sufficient
                if docheckover:
                    cond = self._checkbarover(data, fromcheck=fromcheck,
                                              forcedata=forcedata)

to:

        if consumed:
            self.bar.bupdate(data)  # update new or existing bar

        # if self.bar.isopen and (onedge or (docheckover and checkbarover))
        cond = self.bar.isopen()
        if cond:  # original is and, the 2nd term must also be true
            if not onedge:  # onedge true is sufficient
                if docheckover:
                    cond = self._checkbarover(data, fromcheck=fromcheck,
                                              forcedata=forcedata)
        if consumed:
            data.backwards()  # remove used bar

This is just a quick fix I came up with, it can probably be fixed in a much prettier way.

If you are so confident that the results are wrong, why not just calculate it by hand for some wrong values and check? You can't be sure if they are wrong if you won't do it.

Look at SharpeRatio analyzer code in backtrader and check how many parameters does it have. Any of those parameters will influence the result. Pyfolio Sharpe ratio has to rely on some default parameters as well, they are probably different than bt.SharpeRatio defaults.

Can you add more details please? Standard output - any exceptions thrown? What do you exactly mean by 'this doesn't work'

Seems like you are short selling something which than raises in value by a lot. To simplify analyzing I think, that you should add writer and analyzer (pyfolio analyzer adds a lot of useful output):

cerebro.addwriter(bt.WriterFile, out='OUTPUT_FILE_PATH')
cerebro.addanalyzer(bt.analyzers.PyFolio, _name='pyfolio')

After you add those 2 you will have all positions written to your file, which should be quite easy to analyze when the value went below 0 and what trades did you make before.

Hope that it will help.

I came into a problem while trading a lot of different length data. I am using MomentumOscilator, with a period as a parameter. I am trading big number of different datas, which have different lengths, so sometimes it happens that len(datas[i]) < period for some datas. Currently, I am just filtering the data in the init method, because MomentumOscilator code is throwing an exception while it is instantiating the data. When I checked the code, it actually seems like really simple change to fix this, in backtrader.linebuffer._once method:

        self.preonce(0, self._minperiod - 1)
        self.oncestart(self._minperiod - 1, self._minperiod)
        self.once(self._minperiod, self.buflen())

could be changed to:

        self.preonce(0, self._minperiod - 1)
        self.oncestart(self._minperiod - 1, min(self.buflen(), self._minperiod))
        self.once(self._minperiod, self.buflen())

Next bug related to resampling, I found it while I tried to resample with time_frame = TimeFrame.Minutes with data compression=60 and target compression = 60*24. Problem is located in bactrader.analyzer._get_subday_cmpkey().
Lines:

        dtcmp = dt.replace(hour=ph, minute=pm, second=ps, microsecond=pus)
        dtcmp -= tadjust
        if extradays:
            dt += datetime.timedelta(days=extradays)
        dtkey = dtcmp

        return dtcmp, dtkey

should be changed to:

        if extradays:
            dt += datetime.timedelta(days=extradays)
        dtcmp = dt.replace(hour=ph, minute=pm, second=ps, microsecond=pus)
        dtcmp -= tadjust
        dtkey = dtcmp

        return dtcmp, dtkey

Basically, the extradays adjustment was done too late.

Did some more investigation about it and found out that indeed it is a bug - resampler just don't work at all for timeframe == days. To fix:
The code snippet I placed in previous message:

 self.componly = (self.subdays and
                         data._timeframe == self.p.timeframe and
                         not (self.p.compression % data._compression)
                         )
...
self.componly = self.componly and not self.subdays

should be changed to (add 'not' before self.subdays and remove the last line):

self.componly = (not self.subdays and
                         data._timeframe == self.p.timeframe and
                         not (self.p.compression % data._compression)
                         )

Additionally in resamplefilter.py::Resampler::call() change this:

        if consumed:
            self.bar.bupdate(data)  # update new or existing bar
            data.backwards()  # remove used bar

        # if self.bar.isopen and (onedge or (docheckover and checkbarover))
        cond = self.bar.isopen()
        if cond:  # original is and, the 2nd term must also be true
            if not onedge:  # onedge true is sufficient
                if docheckover:
                    cond = self._checkbarover(data, fromcheck=fromcheck,
                                              forcedata=forcedata)

to:

        if consumed:
            self.bar.bupdate(data)  # update new or existing bar

        # if self.bar.isopen and (onedge or (docheckover and checkbarover))
        cond = self.bar.isopen()
        if cond:  # original is and, the 2nd term must also be true
            if not onedge:  # onedge true is sufficient
                if docheckover:
                    cond = self._checkbarover(data, fromcheck=fromcheck,
                                              forcedata=forcedata)
        if consumed:
            data.backwards()  # remove used bar

This is just a quick fix I came up with, it can probably be fixed in a much prettier way.

Not sure about it yet, but seems like resampler doesn't work for daily timeframe (still investigating if it is sth my code is doing or backtrader bug). Not sure if this is the cause yet, but I think I found a bug in resamplefilter.py::_BaseResampler::init():
Seems like self.componly can never be True:

 self.componly = (self.subdays and
                         data._timeframe == self.p.timeframe and
                         not (self.p.compression % data._compression)
                         )
...
self.componly = self.componly and not self.subdays

In the line:
self.rets[self.data0.datetime.datetime()] = entries
datetime is accessed through self.data0.datetime. It is causing problems when you develop strategies with many datas from which some start earlier than others and you start trading from the moment when first data becomes available.
I think that easiest solution is to just change the line to:
self.rets[self.strategy.datetime.datetime()] = entries

Thanks for they reply, sorry for not answering earlier, but I just forgot that I made this question here. I just wanted to write how did I solve it. I finally ended up with extracting those indicators from outer ones and then adding to strategy indicators using bt.Strategy.addindicator() method in the strategy stop() method.