In this challenge, act as if you were a cross-asset portfolio manager who wants to introduce machine learning techniques to generate insights applicable to an asset allocation strategy. We have selected a list of 19 futures that you can invest in. The list covers all asset classes: blue-chip equity indices, major government bonds and gold.
This is a very exciting challenge which gives you an opportunity to build an all-weather portfolio, switching opportunistically from one asset class / one region to the other to best surf economic / financial cycles using machine learning algorithms. The goal is to outperform the conventional playbook for asset managers which is to invest in equity and fixed income in equal proportion to weather different market conditions. See the benchmark section.
You must build a long-only, fully-invested (and non-leveraged) portfolio. You cannot hold more than 20% in any asset to avoid excessive portfolio concentration. Hence, your exposure to each of the 19 underlying assets should always be in a range of [0;+0.2] and your portfolio total exposure must always be +1.
Your objective is to maximize the annualized return of your portfolio, but also to minimize the drawdown expressed by the minimum 3-year rolling Sharpe ratio. These 2 metrics will be taken into consideration to select the winner(s) along with a robustness score.
Use the sample notebooks to get started:
The Investment Universe is composed of 19 (properly rolled) futures contracts covering all asset classes. The universe can be called using the getTickersGlobalAllocationChallengeV2 function.
Ticker | Name | Asset Class | Currency |
---|---|---|---|
DM | Dow Jones Industrial | Equity | USD |
ES | S&P 500 | Equity | USD |
NQ | NASDAQ 100 | Equity | USD |
VG | EuroStoxx 50 | Equity | EUR |
Z | FTSE 100 | Equity | GBP |
SMix | SMI Swiss Market Index | Equity | CHF |
NX | CME Nikkei 225 | Equity | JPY |
HI | Hang Seng | Equity | HKD |
KM | KOSPI 200 | Equity | KRW |
XP | ASX 200 | Equity | AUD |
TU | US 2-Year T-Note | Fixed Income | USD |
FV | US 5-Year T-Note | Fixed Income | USD |
TY | US 10-Year T-Note | Fixed Income | USD |
US | US Long Bond | Fixed Income | USD |
RX | German 10-Year Bund | Fixed Income | EUR |
G | UK 10-Year Gilt | Fixed Income | GBP |
JB | Japan 10-Year Govt Bond | Fixed Income | JPY |
XM | Australia 10-Year Govt Bond | Fixed Income | AUD |
GC | COMEX Gold | Commodities | USD |
The benchmark will be 50% MSCI World Index Total Return (USD unhedged) and 50% JP Morgan Global Bond Index (USD hedged) with a monthly rebalancing. To check the time series:
Maximum exposure of 20% for each asset at any rebalancing date.
Transaction costs: 1 basis point (bp) of the amount being transacted. As an example, if you sell 10% of S&P 500 and buy 10% of US 10-Year T-Note, the cost will amount to 20% x 1 bp = 0.2 bp.
Transaction frequency: The maximum transaction frequency is daily (no intraday trades). All trades are executed at market close prices.
Transaction signals: Transaction signals must be based on data available at the close-of-business day. Signals are 'executed' the same day following the Lag 0 model convention.
First transaction: First weights have to be generated before/on 2004-01-02 (1 year calibration/training. Rolling/expanding windows are allowed). If returns are NA for a particular asset, the weights assigned to it has to be 0.
Rebalancing penalty: 2 basis points on the total AUM, each time you have a date with a set of weights, it will be considered as a 'rebalancing' day and you will incur the penalty. Please check the example notebook and check the page Rebalancing and transactions costs
The in-sample period runs from 1 Jan 2003 to 31 Dec 2018. To prevent any overfitting of the models, data for 2016-2018 will only be unveiled after your first model submission.
Note: Once your first submission is approved, you can revise it as much as you want.
You must use Notebook to submit your solution. You can only submit one notebook. However, there is no limitation in the number of revisions you can make on a submitted notebook. If the solution follows a theoretical or academic paper, the participants are encouraged to attach the document with the submission, any relevant document (own paper or other document) to explain the approach used will increase the robustness scoring.
Please visit our Submissions page for all details on how to submit.
The strategies will be ranked based on the weighted arithmetic mean of their following two rankings:
Performance Metrics | Metric ranking | Weight |
Normalized Annualized Return | AR | 40% |
Normalized Robustness Rank | RR | 30% |
Normalized Minimum Rolling Sharpe Ratio (*) | MRSR | 30% |
(*) The Minimum Rolling Sharpe Ratio will be calculated over a period of 3 years.
Note:
The total prizes in euros for the winner are:
Tuesday 17, August 2021.
Monday 29, November 2021 midnight NY time.
The only available input data for this competition are: Close price of the 19 tickers listed in the Investment Universe section above.
The data can be called using the getTickersGlobalAllocationChallengeV2function.
The Challenge will be open to any Participants (academic or non-academic) in the world.
Submit early to earn points that will help you win the competition !
Weekly point rewards
Ranking | Points |
1 | 15 |
2 | 11 |
3 | 7 |
4 | 3 |
5 | 1 |
The Challenge will be open to any Participants (academic or non-academic) in the world, but Alphien reserves the right to reject some applicants if terms of the platform are violated. In particular, you must not copy code or intellectual property which is not your own or is not open sourced. You can not participate as a professional if your employer can claim ownership in any part of your work on the Alphien platform; in case of conflict please contact and discuss with the Alphien team. Alphien is open to free scientists who own their intellectual property.