Best posts made by support

@anshul96go Just to give you more context: when you submit a strategy to the contest, the strategy is queued. Depending on the current load, the processing can take some time. If you submit when many other users are submitting, then processing will take more time.

Moreover, when you develop on the notebook, processing is instantaneous because no particular checks are done (well, it depends on the complexity of the strategy). You just run your code and get the result.

If you code it using a single-pass approach the processing will be very fast. If you use our backtesting function (which prevents forward looking), processing will be a bit slower.

When you submit the strategy, however, several sanity checks are run on the strategy. The system checks if allocations are defined for all datapoints, if your strategy is performing looking forward operations, if it stops producing the output before current day, it computes the Sharpe ratio and other statistical indicators, and moreover it computes the correlation of your strategy with all examples we provide, as we do not allow an example to win a contest...so the processing is slower.

For quick development, please refer to the results you get in the notebook, submit the strategy and then it will be ok.

@rezhak21 Hi, the lookback period used in the builtin backtest function is expressed in calendar days, while indicators are computed in trading days. As a rule of thumb, add 2 more days every 5 trading days to take weekends into account.

@spancham Hi, yes, understood. The 10% rule applies to the contest entries and to the prizes (1M USD invested, 500k USD invested, etc, see https://quantiacs.com/contest).

However, strategies can get funded by investors even if they do not win contests. In this case 2 schemes are possible:

1. a fixed management fee, i.e. the quant earns a monthly fixed fee.

2. a performance fee, i.e. the quant earns a monthly performance fee.

It is already happening with quants who submitted systems which developed a long track record. In both cases, there is a mutual agreement between Quantiacs, the investor and the quant on the level of the fees.

Hi.

When you submit the strategy, the evaluator checks the strategy using data isolation and runs the notebook each day during the in-sample period.

This is necessary because there is a very common issue - looking forward.

The evaluator can parallelize this process, but anyway it takes more time.

@spancham Hello, the management fee is order of magnitude of a performance fee. Imagine that you get 1m USD allocated, the system has 10% volatility, a Sharpe ratio of 1 for 1 year, and you make 10% in performance fees per year. Instead of choosing a performance fee (which could be zero in bad months) you could choose at the very beginning a fixed fee. In this case, it would be order of 1k USD per month.

The length of the track record depends on the strategy and many factors, let us say no less than some months, longer time for a larger capacity.

@sheikh Hi Sheikh, we are busy preparing the new contest.

The algorithm you mention actually has a lot of flat phases also in sample, why do you think the problem is an external library?

If you run the system in a notebook and plot the equity chart, do you get the same result or not?

Thank you

@spancham The performance fee is a yearly performance fee as per industry standard. If system makes 100k USD in profits per year (1 Sharpe, volatility 10%, 1M USD invested), then the quant will receive 10k USD per year.

@xiaolan Hello, the transformation is very simple. Let us suppose that you load the data, and select some field, for example the close, from the native xarray structure using:

import qnt.data as qndata
data = qndata.cryptodaily.load_data(min_date="2016-01-01")
close = data.sel(field="close")

You can convert to pandas using:

close_pandaized = close.to_pandas()

Then you can work with pandas. Let us come to the final weight, and let us say they are a pandas dataframe. You should convert them to the final xarray data structure using:

weights = weights_pandaized.unstack().to_xarray()

@magenta-grimer yes, correct, and it does not have to win a contest, we monitor all submitted systems and will contact the quants who wrote interesting systems.

@rezhak21 Hi, correct, you made 5 referrals, we can see that, but none of them made a successful submission to Quantiacs...you will get 15USD per each user who makes at least one submissions which passes our filters. In that case, also the users you referred will get 15USD.

@xiaolan Hello, first of all conversion xarray -> pandas -> xarray is very easy, as we answered you in the other thread you opened.

The basic reason is that xarray works really well with array with an arbitrary number of dimensions. For financial data we have normally 3 dimensions:

• time coordinate
• asset
• field

and xarray works really well with cross-sectional indicators ranging over several assets.

Dear magenta.kabuto, it will be done at the open price in t+1 time.

@anshul96go Hi Anshul, the IS period starts on January 1st 2014. You can use a strategy with the following logic:

1. if Volume/OI is zero, then take this decision:

2. Otherwise, take this decision:

But yous algorithm should produce results also in the period when Volume/OI data were not available.

@laudis
Hello.

It is better to use Linux. It is harder to use multiprocessing in Windows because there is no fork on Windows.

Regards.

@rezhak21 Hi, good question, yes, you can!

Please take a look at the documentation at:

https://quantiacs.com/documentation/en/reference/evaluation.html#stateful-multi-pass-backtesting

When you use multi-pass backtesting you have to define a strategy function which loads the data and returns allocation weights for a fixed point in time. The "default" behavior of the backester does not save the state of the simulation between iterations, normally this choice is safe and fast as it allows for parallel processing.

If you want to preserve the state, you have to pass two arguments to the function: the data and the "state". "state" should be a dictionary containing the variables and the values which should be persistent across iterations. The function should then return allocation weights and the updated state.

Please try the example at:

https://quantiacs.com/documentation/en/reference/evaluation.html#stateful-multi-pass-backtesting

and let us know.

@spancham Currently there is no investor tab, it is part of our roadmap and we are working on that. At the moment we are focusing on improving the software and the data, the new version of Quantiacs is up and running since 3 months only. As soon as the fund is up, we will announce it.

We published a summary of the past 14 contests on the home page, with quant names and allocations made by Quantiacs (own money). Quantiacs has private agreements with investors allocating their money to selected systems, and with quants who developed systems and are getting fees.

If you/your family have enough capital and want to invest in your own algo and bear the risk of downside losses, well, you will be able to do it with Quantiacs (once we start the fund) if you want.

@cyan-gloom Hi, before that, are you sure that your code is not looking forward? There is a shift with a negative value "-1" for the variable "diff", which in turns is used via "signal" for defining "weights". Naively, we would say that the code is looking into the future, correct?

@cespadilla Hello, the Q17 contest has been deployed. We decided to organize a contest very similar to the Q16, but relaxing the long-only condition and allowing quants to short cryptos. You can already play with the new contest and submit, there will be time until April 2022. New data will soon be ready with the Q18.

@captain-nidoran We would like to add more assets in the next days, without removing any of the current assets.

Please note that all the past submissions you and other quants made before the data update are still running on the previous dataset, and they are eligible for the Q15 as they were submitted.

So adding assets is never affecting existing submissions, as each submission is bound to the dataset used for developing and submitting.

But if you implemented a stateful strategy, the evaluator can't parallelize the checking. It will take much more time. You can see the time of the one-day evaluation in the log and estimate how long it will take.