Quantopian closing

[kumar]

Marco/P123 Owner.

The manual stock pick quarterly and yearly holding paid subscription model will be one of p123 REVENUE generating model for the manual stock picking experts and p123.
Like P123 did in March 2013 launching of automated designer model for the subscriber.

===========================
I am expecting January 2021 will have a new world of manual stock picking models with proven experts initially, it could have free subscription during the starting time to make the the subscriber comfortable
with the free subscription for few months.

Successful 10-20 Handful of stock picking system using all the wisdom is Low risk investment.

P123 - is a portfolio tracker - it could be automated designer model ; and it could be manual stock pick for every quarter or every year;
If p123 focus on this business; It will have additional stock picking experts subscription and membership subscription to generate revenue;

My performance for 20 stocks for 1 year holding; as part of the p123 yearly competition since 2017.
2017 35%+ 1st place
2019 40%+ 2nd place
2020 60%+ 1st place
2018 reasonable down percentage with market down

My models are very liquid
will speak for my stock picking skills as

• One of the best example for subscriber to follow the model to get profits better than bench mark - yearly once buy activity
• One of the best example for the stock picking expert to demonstrate their skill thru performance.
• One of the best way p123 to be come famous with many successful manual stock pick strategies which out perform bench mark significantly.
• It will attract new subscriber to follow
• It will bring more revenue to P123 by new subscriber who follow very successful p123 stock picking models and commision from the manual stock picking membership fees.

Thanks,
Kumar

[abwillingham]

I would be interested in Europe and World data, however I don’t have any experience trading in those markets.
I recently researched how one could trade South Korean stocks and came to the conclusion that they make it nearly impossible for an individual outside trader to be successful there.

What does the rest of the world look like? I assume trading in Europe is straight forward.

[geov]

Sorry about the mixed messages. We had decided a few months ago to go ahead and purchase world data, and finance the whole thing ourselves. We were very close to pulling the trigger. Several things have happened that made this a high risk proposition, and potentially a money losing / cash draining project:

• Huge amount of uncertainty as the world is headed for another lockdown. The first thing people cancel to conserve cash is these types of memberships
• Slowdown of the entire sector : quantopian closing for example, but others are complaining
• Performance! people in the end want to make money but quant/value strategies have been underperforming. Buying an index was a wise move for the past 5 years. Buying the QQQ would have been brilliant.
• Being the only ones to have S&P data was a real differentiator. Having to switch to FactSet has not been good to us.

So we decided to hunker down for a bit. Maintaining our cash reserves during this uncertain times is more prudent.

I agree with Marco. This is not the time to embark on an uncertain venture to bring world data to P123.
Perhaps the current platform can be improved instead, like providing the promised Weight Formula for books, for example.

[marco]

Money talks.

We’ll run the pledge campaign again to see if we can raise the cash. Doesn’t cost anyone anything.

The amount we’re aiming covers the data and the labor.

But yeah, if we hunker down, another thing we can do is take a hard look at low hanging fruits. Review outstanding requests. Perhaps look what else is in the database that we’re already paying for that can be of use.

For example I just found that we get the timestamp of the earnings press release. So we do know if the earnings press release was before , during or after market. With this data we could add new factors like performance 1w, 2w, 1m pre/post earnings release. This would facilitate doing PEAD studies - Post Earnings Announcement Drift . More on this soon.

Thanks

[jmh]

What does the rest of the world look like? I assume trading in Europe is straight forward.

Something to take into account → it looks like the cost of trading in Europe is much higher (commissions)

See for example for IB → Commissions | Interactive Brokers LLC

I have not done the maths but it will impact strategies even with moderate turnover [250%-400%]

Jerome

[geov]

Jerome,
What about taxation on trading profits? There may be automatic tax withholding rules in foreign countries when as a non-resident you sell a position at a profit. What about taxation of dividends? Also what are the reporting requirements. Does one have to file a tax return in the foreign country if one is a US person? You may have to file overseas additionally to filing in the US. Also a FBAR filing is required in the USA for foreign bank and brokerage accounts totaling in excess of $10,000 anytime during the year. Failing to do so can land you with a $100,000 fine or 10 years in jail.

These are all issues that one should know before trading foreign stocks on foreign stock exchanges and should be investigated. If you have to hire a tax lawyer who is familiar with all this it will cost plenty.

[philjoe]

only tax problem in europe is buying on LSE afaik

[Quantonomics]

Perhaps look what else is in the database that we’re already paying for that can be of use.

You have access to Canadian ETFs data and you aren’t doing anything about it. Is there a way you could make it available for ranking and strategies the same way as US ETFs? That would be amazing, thanks!

[jmh]

Hi Georg,

I am based in Europe and invest in the US. As far as I have experienced there are taxation treaties to avoid you being taxed twice on the same thing. This is also made so that you end-up being taxed under the regime of the country where you live.

For example, the US systematically applies a dividend withholding tax at the point of the dividend payment (normally 30% - less if you complete a form W8-BEN). There is therefore a part of the dividends that I never see hitting my portfolio as it goes straight to the IRS (thereby slightly impacting my ports vs model as I cannot reinvest what went straight to the IRS). However, when comes the time to do my taxes here, the amount withheld by the IRS is subtracted from what I owe to my local tax authority - effectively meaning that I am taxed the same way as if the dividends had been received from European stocks and traded on European Exchanges.
Similarly capital gains tax apply under your local rule e.g. there is here no such thing as short-term or long-term trades even on US stocks.

I assume the reverse situation (living in US, trading in Europe) is similar as in you will bu subject to whatever tax rule applies to you in the US even if you invest in Europe.

Back to my main point, I remember paying £10 transaction fee to buy an ETF on the LSE. Commissions are much higher in Europe and if one has a port with 800% turnover, one will be in for a surprise to the downside…

JMH

[judgetrade]

yep, understandable that P123 does not want to add risk to the business modell, keep up the level you have and go step by step, fine by me.

And if the need for EU Data is really that high, then a new fund raising round will show if the deman is high enough.

Best Regardsa

Andreas

[SpacemanJones]

For example I just found that we get the timestamp of the earnings press release. So we do know if the earnings press release was before , during or after market. With this data we could add new factors like performance 1w, 2w, 1m pre/post earnings release. This would facilitate doing PEAD studies - Post Earnings Announcement Drift . More on this soon.

Interesting. I think event driven studies like this could be useful in the context of models. There’s quite a bit of strange price behavior around earnings that I don’t understand - and I’d guess different models might present different behaviors. An event flag would provide an interesting way to study the expectation.

[Jrinne]

@duckruck

In my experience on non-US quantitative platforms, excessive skill requirements do significantly reduce the quality of the strategies people create. On these platforms, people even believe that many programs that perform significantly worse than the widely known one-factor strategies are very good strategies (and then the platform managers get angry when I point this out).

That doesn’t mean that more advanced features are useless, but too high a bar might hinder the community.

Duckruck,

TL;DR: I do not think random forests are too complex or too advanced for most P123 members. I agree linear models work well and have advantages. But the Python code required for a random forest is trivial for most P123 members now–which opens up the ability to investigate non-linear methods at P123.

IMHO, there is absolutely no reason, a priori, to think a linear model is best. Not when that conjecture can be tested.

Random forest are doing well for me, as is Elastic net regression. I do not think, you could go wrong with some of the linear methods you have suggested in the forum.

Random forests are SIMPLE for anyone wanting to investigate non-linear models. They are easier than Principle Component Analysis for me when you get right down to it.

A grid-search of the ‘min_samples_leaf’ and ‘max_features’ with the code below is all one needs to determine the best hyperparameters for a random forest. One could consider recursive exclusion of some of the features after that depending on the confidence you have in the factors you have selected.

After that, you are done with nothing left to do, and you will have to look for new factors if you want to keep looking into new ideas.

Random forests are not prone to overfitting. There are only 2 hyper parameters that need to be addressed with little coding required to do it. My experience suggests larger numbers should be used for ‘min_samples_leaf’ as reflected in the code below :

from sklearn.model_selection import GridSearchCV
from sklearn.ensemble import RandomForestRegressor

Define the parameter grid

param_grid = {
‘min_samples_leaf’: [1000, 2000, 4000, 8000],
‘max_features’: [‘0.3’', ‘sqrt’, ‘log2’, None]
}

Create a random forest regressor object

rf = RandomForestRegressor(random_state=42)

Create the grid search object

grid_search = GridSearchCV(estimator=rf, param_grid=param_grid,
cv=5, n_jobs=-1, verbose=2, scoring=‘r2’)

Fit the grid search to the data

grid_search.fit(X_train, y_train)

Get the best parameters

best_params = grid_search.best_params_

Train and predict using the best parameters

best_rf = RandomForestRegressor(min_samples_leaf=best_params[‘min_samples_leaf’],
max_features=best_params[‘max_features’],
random_state=42)
best_rf.fit(X_train, y_train)
predictions = best_rf.predict(X_test)

Jim

[ustonapc]

Duckruck,

I agree that the risk-adjusted performance of merger-arb strategies looks attractive. (pls see charts below) However, there are two main issues of applying merger-arb strategies in P123.

Professional Merger-arb managers usually invest long in the acquiree and at times short the acquirer to hedge the risk. There are no filters to seperate the acquiree and the acquirer in P123. I have a suggestion for those interested to try on P123 which is to buy the companies with prices rising in the previous month (usually the acquiree) and short the companies with negative price performance in the previous month (usually the acquirer) to hedge the risk.

Another problem is using monthly rebalancing since the price of the acquiree can dropped very significantly if there is a sudden regulatory problem with the merger companies which is why MNA etf and MERFX mutual fund performance is suboptimal.

Regards
James

[Jrinne]

@duckruck

After all, even the simple strategy of equal weighting academic factors is a difficult hurdle for many to cross. These things are simple, but not easy.

Duckruck,

I do not know much about mergers and I am all about the methods (not the particular features as much).

In that regard, I wonder if you use Feature Importances in a random forest or a boosting program to help you select features.

And I should probably ask about Lasso Regression and/or Recursive Elimination (which can be automated with Sklearn as you know) as methods for selecting factors.

I am not recommending a method, I would just be interested in what you might be doing in this regard if it is a topic of interest to you (and want to discuss it in the forum).

Jim

[Jrinne]

Nice! Thank you very much.