Skip to content
Taoshi Documentation

Closed Positions

The objective of subnet 8 is to procure usable and high quality trading signals. When we first launched PTN, we just looked at the returns from closed miner positions. Miners found a number of ways to skirt the rules laid out by the system, from taking high leverage YOLO positions to leaving losing positions open. To better manage how copy traders are likely to experience the trades taken by the miners, we needed to design a system capable of addressing losses and risk from positions in real time. We created and introduced a system of unrealized returns to track miner performance, which looks at their portfolio value change over a thirty day period to determine their ranking among other miners. We also introduced risk adjustment based on max drawdown, which helps manage overall portfolio level risk in realtime. This was a critical step towards normalizing miner behavior and making it usable in a network model.

With the current structure, miners can continue tracking positive portfolio value increase without closing positions, increasing their score in the system. The intention with this design is to allow customers of PTN to attach themselves to top miner signals at any time, copying the same leverage from the miner. Note: unrealized returns mean any portfolio value movement measured by the miner over the lookback period, regardless of if the positions are closed or not. Realized returns are for the purposes of this proposal defined as closed positions and open losing positions.

Problem Statement

In theory, this structure should work to incentivize sustainable strategies. If buy/hold strategies are not able to sustainably bring in returns, then they should not be competitive with strategies which are more responsive to market movement and thus should not be in the top 10. This theory relies on a long enough lookback window, to force miners without a strategy to be statistically pruned from the network as they continue to make guesses about market movement. It is also a function of the maximum leverage available to the miners, as over leveraged miners are able to more easily jump to the top of long term returns. If the expected value cost of registering a miner is lower than the potential payout from a high leverage trade, this behavior would likely be prolific.

We have no confidence in buy / hold miners to continue to trade well without an indication that there is some level of complexity from their strategy. Institutional investors using PTN are going to look for the same metrics they use to evaluate other traders outside of PTN to determine the value of miners. These metrics are based on closed positions, and are not available to us with our current structure. We will change from evaluating unrealized returns to realized returns to better align with the metrics used by institutional investors.

Proposed Changes

There are a number of system level changes that we need to make to transition from unrealized returns to realized returns:

  1. Scoring
  2. Challenge Period
  3. Penalties / Consistency
  4. Hyperparameter Tweaking

Scoring

The first is to change the way we evaluate miner performance. We will no longer look at the difference between the current portfolio value and the portfolio value at the start of the lookback window. Instead, we will look at all positions which were opened and closed within the lookback period, and evaluate the returns from those positions. Of note, we will use closed positions and open positions with negative return to calculate the current realized returns for the miner. This should discourage miners from leaving losing positions open, as they will be penalized for the loss in their realized returns.

Additionally, we will include any open positions with negative return as part of the miner’s realized returns, regardless of when the position was opened. This means that losing positions opened from over one month ago will negatively impact the miner’s realized returns - to remove this from their realized returns, the miners should simply close the position. Once they close the position it will no longer be tracked in their current realized returns, as we require both the open and close times to be within the lookback window.

This logic will apply to both short term returns and long term returns. Short term returns are positions which were opened in the prior 30 days and closed recently. Long term returns are any positions which were opened and closed in the prior 30 days.

Challenge Period

The challenge period has been plagued with legacy rules making it hard to interpret for new miners and inconsistent with the rest of the system. Here are the current rules:

  1. 2% Total Return
  2. -3.08E-8 Time Averaged Sortino
  3. 12 Volume Minimum Checkpoints

Time averaged sortino and volume minimum checkpoints are a major point of confusion, and must be removed as they are not even used for scoring in PTN.

We will update the challenge period with the following rules:

  1. Realized Returns: 2% total return from closed positions and any open positions in loss.
  2. Realized Returns: A single position should not account for more than 25% of the total returns.
  3. Unrealized Returns: A single day of trading should not represent more than 20% of the total returns.
  4. Unrealized Returns: Portfolio value should never experience a drawdown of 5% or more.

Penalties / Consistency

We are going to keep a blend of unrealized and realized penalties to evaluate the quality of the miner, to avoid scenarios where the miner masks a single instance of success as sustained benefit. Here is a summary of the penalties we will likely use:

  1. Closed Positions: Ratio between max positional return and total return.
  2. Portfolio Value Movement: Ratio between max daily portfolio value change and total portfolio value change.
  3. Portfolio Value Movement: Max Drawdown.

The last two portfolio value movement penalties are already incorporated into PTN via our perf ledger system. These will remain or receive marginal changes alongside the introduction of the new max return ratio on a single position.

Hyperparameter Tweaking

We will continue to prioritize long term returns, and short term returns will play a role. To avoid the scenario where short term miner returns fluctuate substantially as miners close positions, we propose an increase to the lookback window on short term returns, to 5 days from 3 days.

Additional Information

  • Any miner currently in challenge period may request a clean slate if they choose. Their existing positions will be wiped and their entry time will be reset.
  • We will also look to extending the lookback window to 90 days, which will be rolled out with this release. Details on this change will be released in another proposal tomorrow.

Timeline

Proposal Delivery: August 5, 2024

PR: August 12, 2024

Implementation: August 23, 2024

Pull Request