The Robot Index (iRobot) - Updates & Final Backtest before DG1

[Monportefeuille]

IIP-XX
title : Robot Index (iRobot)
status : Proposed
author : bloom.fi DAO represented by @Monportefeuille
reviewed by : @Thomas_Hepner
created : 2021-07-16

Summary

This post is the formalization of the original PMI (Post-Modern Portfolio Theory Index) proposal drafted 3 months ago. It aims at summarizing the progress and changes made since then, paving the way for a DG1 vote.

• Mission & Purpose :
  The Robot Index (iRobot) aims to become the 1st decentralized Robo-Advisor, automatically optimizing the composition of a sector-agnostic basket of crypto assets, by selecting and rebalancing its underlying components based on their annualized risk-adjusted returns.

• Thesis :
  The core of Index Coop’s mission is to make crypto investing simple, accessible & safe.
  Unfortunately, and despite the potential offered by market-capitalization weighted benchmarks, building a diversified portfolio with long-lasting performance while minimizing its downside deviation is not always immediate for retail investors.
  This is particularly true for those entering the market in periods of high volatility, and hoping to outperform the top market caps. With generally little trading or portfolio management experience, a natural bias when facing a screen full of numbers and obscure token names is often to place bids on random projects, or based on external advice or crowd sentiment.
  The Robot Index aims at taking this difficulty out by automatically picking up the assets with the most growth potential and the least downside volatility.

• Product Differentiation :
  There is no equivalent product elsewhere on the market. Thanks to its unique characteristics, the Robot Index also fills an interesting gap in the Index Coop’s product palette by unlocking the potential of quantitative trading techniques to supplement the broad sector exposure thesis.

• Market Opportunity & Potential Size :
  1/ According to Statista, assets under management in the TradFi Robo-Advisors segment are projected to reach $1.4 trillions this year (~ 1.6% of the world’s top 50 asset management firms AUM). This market is expected to show an average annual growth rate of ~ 19% in the coming years.
  2/ Assuming decentralized Robo-Advisors capture an equivalent portion of the
  total value - currently - locked in DeFi results in a potential market size of ~ $880m.
  We believe the Index Coop could seize a significant share of this opportunity thanks to the 1st mover advantage that the Robot Index is offering.
  3/ Several major players in this sector such as Wealthfront or Betterment as well as new actors like Makara, are looking to attract digitally native generations with an ETF-like approach to crypto offering. Before its official launch, Makara already had 20000 beta testers with several million $ under management while its offer is currently limited to 3 “baskets” (“Blue Chips”, “Inflation Hedge” & “Decentralized Finance”).

Evolutions since the original PMI proposal

• Community sentiment :
  The PMI proposal has gathered a large majority of positive votes (88%) FOR the progression towards a DG1 vote.

• Product Working Group feedback :
  A pre-DG1 product priorization review was conducted by the Product Working Group (thanks to @overanalyser & @Metfanmike for their guidance on this). Based on the revised scoring framework, the Robot Index achieved an initial score of 1.38.
  The diversification and research automation have been identified as the potential strongest growth drivers, while the complexity of the methodology justifies a higher revenue potential.
  The adjustment of the rebalancing period to a monthly basis for the initial launch would be a further help in reducing the operational overhead with a potentially beneficial effect on the above score.

• Engineering Working Group feedback :
  After a 1st discussion on the methodology’s technical implications with @dylan, it is envisioned to use the same interface for the rebalancing process as MVI or DPI for iRobot’s launch. It was also suggested that a bi-weekly rebalancing cadence would incur too much overhead in parametrization and actual trade monitoring, at least for the launch : although the original recommendation for this product is a bi-weekly rebalance to manage risk adjusted returns as efficiently as possible, the latest backtest data show that this would still be acceptable (read below the “Backtest & Bear Market Stories” …).
  The automatic calculation of the metrics underlying this methodology directly on-chain is theoretically possible and could be an option for the future, but Set Protocol is currently not set up to enable this.

• Author Background & Commitment :
  Since early June the bloom.fi DAO has opened its doors to the broader English speaking community, with the aim to offer curated crypto / DeFi market research and analytics.
  I personally believe in the power of a collective intelligence to share & process the data generated by the most advanced research platforms, extract signals from the noise & provide a new type of consultancy / curation service to the industry.
  As an aspiring community methodologist, my intention is to build on the Robot Index to cement a strong connection with the Index Coop and bootstrap bloom.fi’s development in this new direction, including the recruitment of DAO members to support iRobot’s inception & development.

• Narrative :
  Thanks to the input of various community members ( @BigSky7, @lavi, @mrvls_brkfst) we’re confident that we’ve found a positioning which will enable us to communicate around this complex strategy in a simple and exciting way.
  The Robot Index narrative unleashes a lot of marketing potential directed at retail investors, looking for diversification & shorter term returns in addition to BED, DPI or MVI - without the hassle of managing volatility themselves : memes, embodiments of the product, for example via dedicated (& ready to fire !) Twitter or Discord accounts …

Backtest & bear market stories

The 1st methodology backtest shared back in May was covering the period from 1st December 2020 until 1st March 2021 (see the original PMI discussion for details).
Obviously we were still in the middle of a full blown bull market, and a lot has changed since then … It makes sense to look at iRobot’s performance against the same benchmarks (BTC, ETH, DPI and BED) but in a totally different context, also bearing in mind the initial constraint of a monthly rebalancing cadence to limit the product’s operational overhead.

• How would iRobot’s draft portfolio look like 3 months down the line ?

Illustrated below is the composition of the Robot Index as of 1st July 2021, identified during a methodology backtest covering the period between 1st May & 31st July 2021 :

It is noticeably different from the composition proposed in the original PMI post, for 2 reasons :
1/ The ranking of tokens constantly evolves with the market and liquidity conditions.
2/ The list of tokens for which we’re able to process the annualized Sortino ratio has expanded significantly, from around 200 at the beginning of this year to more than 600 today ! This has been facilitated by the automation of our tools and the development of a dedicated script in Trading View .

• How would this composition have changed over the course of the backtest ?

Illustrated below is the evolution of iRobot’s composition from 1st May (inception) until 1st July (2nd and last rebalancing of the considered period):

Despite the global market volatility, the selection of tokens composing the Robot Index would have stayed fairly consistent. The biggest rebalancing operations, notably 1 month after inception (-5.6% for LON, -3.5% for RLY & +9% for QUICK), would have focused on tokens offering the deepest liquidity (pools ranging from $20m to $30m) without compromising the methodology’s principles nor the index performance.
This and the 2 observations below confirm that, in the current market conditions, on-chain liquidity needs to be considered on a case-by-case basis during the index determination phase :
1/ While pushing the maximum number of components above 15 worked well in a bull trend to minimize the ratio trade size / token liquidity, it would have had a rather detrimental effects on iRobot’s performance in this new context : possible, but not ideal.
2/ Given the significant liquidity discrepancies between all candidates tokens, an “MVI-like” liquidity-weighting would have increased even more the composition bias towards QUICK.

• How would this portfolio have performed since the beginning of the crypto market turmoil mid May ?

Illustrated below are iRobot’s average performance metrics from :

• 1st to 31st May
• 1st to 30th June
• 1st to 31st July

Again, the focus has been put explicitly on this period to simulate the worst possible market conditions for the product launch :

The bigger the average daily return and the smaller the standard deviation of negative returns, the better the Sortino ratio (see this article for a detailed mathematical definition).
In TradFi, the Sortino ratio has established itself as a more accurate indicator than its counterpart, the Sharpe ratio, to track a portfolio’s risk-adjusted return.

Here is iRobot’s annualized Sortino ratio compared to its benchmarks for each sub-period :

While May has been challenging from all point of views, this data confirm that iRobot would have been able to absorb similar levels of downside volatility as DPI, while overperforming the majority of its benchmarks as soon as June.

This trend has clearly accelerated since mid July, in parallel with the global market recovery, exacerbating the upside potential of iRobot’s components such as AXS, WOO, QUICK or ERSDL.

The computation of each asset’s cumulative returns confirms iRobot’s explosive growth in the last 2 weeks (+72% !):

With this in mind, here is how much an initial 1000$ investment would have evolved over the course of this 3 months backtest :

While ETH benefitted from an impressive rally during the 1st 2 weeks of May, a buy and hold strategy applied to iRobot would still have outperformed DPI, BTC or BED.

To conclude this analysis, here is the global Sortino ratio ranking based on the entire dataset covering all days between 1st May and 31st July :

Over the last quarter, the Robot Index and ETH are the only assets to feature a positive Sortino ratio, nearly identical in absolute value. While this performance could have been further improved by switching from a monthly to a bi-weekly rebalancing cadence, the best is yet to come !

Methodology recap

1/ Token inclusion criteria :

• The token must be available on the Ethereum blockchain. This is subject to change if the Set Protocol infrastructure allows multi-chain indexes in the future.
• The token must have at least 1 month of price and liquidity history : from our previous research, this is the minimum amount of time required for the annualized Sortino ratio to stabilize in a plausible range - while allowing to capture the upside potential of the token.
• The token must have sufficient DEX liquidity to support inclusion (currently estimated minimum threshold : $4M liquidity pool). If a token has insufficient liquidity, it will be removed from the index during the determination phase.
• The token must be emitted by a project considered to be building a useful protocol or product : projects with Ponzi characteristics, stablecoins, tokens with no utility or existing primarily for entertainment (meme coins for example) will not be included.
• An independent security audit should have been performed on the protocol and results reviewed by the product methodologist. In the eventuality that no audit has been performed, the methodologist applies subjective judgement of the protocol based on assessment of the criteria above and communications with the team.
• In the event of a security issue the methodologist will work with the project team to understand the issue and any effects to iRobot holdings. The team is expected to provide users of the protocol with a reliable solution and adequate documentation to ensure transparency about any incidents.

2/ Index weights calculation :

• The Robot Index composition should include minimum 10 / maximum 15 tokens in order to capture uptrend potential and dilute risk.
• The index will be maintained monthly at launch, subject to increase to a bi-weekly cadence as engineering constraints & operational overhead becomes less limiting factors.
• During the final week of the month, the eligible tokens will be ranked by annualized Sortino ratio and weighted in proportion of their respective scores with respect to the overall ratio : this is the determination phase.
• Following publication of the determination phase outcome, the index composition will change to the new weights during the 1st week of the following month.

3/ Fees :
Fees will be determined between the DG1 and DG2 vote and before the subsequent product launch.

Voting

Decision Gate 1: TBD

Decision Gate 2: TBD

• Update 2021-07-19 :
  – Mentioned @Thomas_Hepner () as IIP reviewer

• Backtest Update 2021-07-23 :
  – Reassessed liquidity history for all candidate tokens all the way back to 1st May '21 replaced DEUS & UNI by INJ & MUSE in initial index composition to maximize Sortino ratio.
  – Reviewed subsequent rebalancing operations and iRobot composition on 1st June & 1st July '21.
  – Updated average daily return & standard deviation of negative returns with the new data. Also added annualized Sortino ratio for each sub-period.
  – Upgraded iRobot backtesting spreadsheet to take missing days into account in the global Sortino ratio and the cumulative returns calculation.
  – Added cumulative returns for all assets.

• Final Backtest Update 2021-07-31 :
  – Updated all graphs with the data covering the last missing period (23rd > 31st July).
  – Added Value tracking of 1000$ Initial Investment.

[BigSky7]

Wow - blown away by the quality of proposals over the past few days. Index Coop’s fall product launch schedule is shaping up to be both extremely exciting and innovative. We are truly building a market leading suite of products.

I can see all the methodologies currently proposed finding clear product market fit and driving innovation for DeFi as a whole. It seems clear that indexes may play an even bigger role in DeFi than they do in tradfi.

I’m incredibly bullish on this full line up and hope we can launch as many of these as possible. It seems clear that some of the extremely tight engineering constraints faced by our community over the past few months are starting to lift - opening the gates for even more product launches.

Excited to dive further into this proposal and get $iRobot launched.

[Ahuja]

Hi bloom.fi

Could you clarify if the average daily return in the chart is a simple average or geometric average? Also, it may be helpful to include the cumulative returns of the various products from start to the end of the test period.

Would you please also clarify if the annualized sortino ratio was obtained by annualizing the average daily return and standard deviations or monthly values? I bring this up because daily fluctuations can be very noisy and can result in undue penalty.

Thank you!

[Monportefeuille]

Hey @Ahuja,

Thanks for raising these interesting points !

1/ The average daily return in the chart is a simple average over the considered period. It aims to illustrate variations in price of the corresponding asset or basket of assets from one day to another, without taking into account any notion of compounding.

2/ Adding the cumulative returns to the graph is a good idea, I’m working on it and will update the graphs accordingly before DG1.

3/ The annualized Sortino ratio was obtained by annualizing the average daily return and standard deviations. In order to be as transparent as possible and include more data from the month of July than what was taken into account here, I plan to release an update of the above backtest results just before DG1.

I hope I was able to answer your questions, let me know if you need anything else.

[Monportefeuille]

Backtest Update 2021-07-23 :
– Reassessed liquidity history for all candidate tokens all the way back to 1st May '21 replaced DEUS & UNI by INJ & MUSE in initial index composition to maximize Sortino ratio.
– Reviewed subsequent rebalancing operations and iRobot composition on 1st June & 1st July '21.
– Updated average daily return & standard deviation of negative returns with the new data. Also added annualized Sortino ratio for each sub-period.
– Upgraded iRobot backtesting spreadsheet to take missing days into account in the global Sortino ratio and the cumulative returns calculation.
– Added cumulative returns for all assets.

[Monportefeuille]

Dear owls,

After :

• 1.5 years of Sortino ratio scanning for several hundreds of tokens …
• 6 months of exploration and discussion within the Coop to assess the relevance of this index and build confidence into its potential as a product …

We’re slowly but surely coming to the end of the current Robot Index backtest, covering the period from 1st May to 31st July’21 … hence approaching the ideal point in time to submit iRobot to a DG1 vote !

The last update will be published at the end of this week, so that the IIP can be officialized. Let’s not rush any conclusion till then, but the results so far are pretty strong … and the market conditions we’ve been through over the last 3 months give a good example of where this methodology can make a difference !

Here is a thread posted on Twitter by the Robot Index itself if you want to learn more

iRobot

@gregdocter @overanalyser shall we set up a community call towards the end of next week, ideally on Friday 6th August ? Can we also schedule the DG1 vote to begin following the call ?

Looking forward to unveil the final backtest results, and above all to answer your last interrogations before deciding on this product’s future ! (thanks for sharing questions beforehand to enable for optimal preparation)

[anon10525910]

@overanalyser and @puniaviision are the best folks for coordinating product community calls!

Acknowledging that ^ is a recent clarification, as I had coordinated 1 or 2 in the past.

[overanalyser]

DM sent to @Monportefeuille on Discord

[overanalyser]

OK, the community call will be on Friday 6th at 16:00 UTC

iROBOT community call
Friday, August 6 · 16:00 UTC
Google Meet joining info
Video call link: https://meet.google.com/iwk-kseq-oib

[Monportefeuille]

Thanks for setting this up @overanalyser , great opportunity - looking forward to it !

[Monportefeuille]

Final Backtest Update 2021-07-31 :
– Updated all graphs with the data covering the last missing period (23rd > 31st July).
– Added Value tracking of 1000$ Initial Investment.

Ready to formalize this IIP with an official number and a Gitbook entry @anon10525910 @puniaviision @dylan !

[Pepperoni_Joe]

Hi @Monportefeuille

Thanks for delivering the Community Call on iRobot earlier

With my IIP editorial glasses on I notice this IIP is not in our standard format. Would you be able to update it in line with this template?

Also noting that there is a max character count on Snapshot. Unhelpfully it doesn’t say specifically what this is, but could I suggest you provide a slimmed-down version of this document in the standard IIP template so we can progress iRobot to DG1.

[Monportefeuille]

Hey @Pepperoni_Joe sure, thanks for pointing me towards this template, working on it as we speak .

Here is what I would like to do :

• Rename this post as “The Robot Index (iRobot) - Updates & Final Backtest before DG1” so we can keep a trace of the discussions and the full content that was shared here before going into the vote.
• Collect the essence of this document in a separate post, compatible with the standard IIP template and Snapshot limitations.

I hope this works out for you, any issue let me know !

[Pepperoni_Joe]

That sounds great!

Thanks for moving so quickly on this @Monportefeuille.

[Monportefeuille]

Hi @Pepperoni_Joe, the IIP compatible / slimmed down version of this post is nearly ready. I’m finalizing the last details and will hopefully be able to post it later today (CEST).

[Monportefeuille]

Hey @Pepperoni_Joe, just got the proper IIP out ! Thanks in advance for assigning a number to it and queuing it for DG1 ! Let me know if you need anything else

[rootulp]

Thanks for the detailed IIP Monportefeuille. I would love to see a Wealthfront for crypto.

Sorry for the naive question but were the 15 tokens chosen for the backtest the top 15 tokens when evaluated based on Sortino ratio? Are large market cap tokens intentionally excluded?

@Monportefeuille

The token must be emitted by a project considered to be building a useful protocol or product : projects with Ponzi characteristics, stablecoins, tokens with no utility or existing primarily for entertainment (meme coins for example) will not be included.

I agree with this intention but I’m not sure how it will be enforced. Are we trusting the methodologist here to make a subjective decision on whether a token is for a useful protocol?

@Monportefeuille

• The Robot Index composition should include minimum 10 / maximum 15 tokens in order to capture uptrend potential and dilute risk.

Out of curiosity, is there any data for why the 10 - 15 range was chosen?

[Monportefeuille]

Hey @rootulp thanks for your message, and sorry for the slightly delayed feedback

I don’t think your question is naive at all : if this complex methodology requires further clarification or explanation, delivering it to the community as efficiently and transparently as possible is my highest priority.

1/ As you intuited already, the draft index composition that came out of the backtest was based on Sortino ratio evaluation + on-chain liquidity analysis.
2/ There is no specific criteria to intentionally exclude large market caps, for example $YFI is in the top 100 ($1.4 billions market cap) and was included based on its outstanding risk-adjusted returns.
3/ The range of 10 to 15 tokens is also the outcome of the backtesting work, and stands out as the best compromise between numerous parameters :

• With the way weights are calculated, including less than 10 components can lead to bigger weight discrepancies, hence frictions with the management of on-chain liquidity and be detrimental to iRobot’s performance on the longer run.
• Including more than 15 increases the maintenance costs (transaction fees) and can also be detrimental to performance in acute bear phases.

Finally, the decision whether a token can be included or not in iRobot’s composition based on its utility will be based on extensive market research and communicated transparently during the determination phase.