Predictoor: Right-Size Staking in Predictions

How much should a predictoor stake, to maximize revenue and bound risk?

Abstract

This article is for predictoors — people running prediction bots in Ocean Predictoor ecosystem.

If you stake more, you can earn more, but only to a point. Earnings are bounded by sales of the feeds. So if you stake too much, you will lose $.

This article provide equations that bound how much to stake, along with general tips.

1. Introduction

Ocean Predictoor offers data feeds predicting the price movements of the top 10 cryptocurrencies by market cap during the next 5 and 60 minute intervals. Data scientists can run a predictoor bot that submits forecasts for these tokens to Predictoor contracts, and earn rewards.

Rewards are financed by sales of the feeds. These are (i) organic purchases by traders and others, (ii) purchases from Predictoor Data Farming OCEAN side, and (iii) rewards from Predictoor DF ROSE side, which are pro-rata to the other sales.

Rewards are affected by:

• Amount of sales. Sales↑ means rewards↑.
• Prediction accuracy. Accuracy↑ means rewards↑.
• Amount of stake. Stake↑ means potential for more reward, bounded by prediction feed sales; more on this later.

Given all these factors, much should a predictoor stake, to maximize revenue and bound risk? The article provides some answers to that question.

The rest of this article is organized as follows. Section 2 gives an equation for payout as a function of such factors. Section 3 has formulae for upper-bound stake amounts; the models have assumptions but are helpful in practice. Section 4 provides general tips. Section 5 concludes.

2. Payout Equation

For a given predictoor’s prediction: if the prediction is wrong, stake is slashed, i.e. they lose the stake. If the prediction is correct, they receive a payout.

Under a correct prediction, payout is:

The initial terms specify that the rewards are distributed only among those who predicted correctly. In order to maximize rewards, participants must not only be accurate in their predictions but also stake an amount reflecting their confidence in these predictions.

If all participants stake the same amount, rewards are equally divided among those who predicted correctly. However, if a participant makes a correct prediction but stakes less than others, their reward is smaller. Conversely, placing a high stake on an incorrect prediction results in the loss of the entire stake.

3. How Much To Stake?

Given these factors, it’s crucial to carefully decide how much to stake. To better understand this, let’s examine some scenarios from simple to complex.

3.1 How Much to Stake: Single Predictoor Scenario

The objective is to determine the appropriate amount to stak.

Let’s start by assuming there is only one predictoor in the system . Therefore we can ignore factors related to other predictoors’ stakes. We only have to consider reward amount and accuracy.

Let’s assume that their classifier’s accuracy is 50%, i.e. random, making correct predictions half of the time.

Then, expected payout E[Payout] is:

And, expected return E[R] is:

3.1 How Much to Stake: Multiple Predictoors Scenario

Assume now that other predictoors enter the game. The main modification to the equations is that the % of total rewards for a particular predictor with arbitrary accuracy Acc, is weighted by the amount of stake.

Then, expected payout is:

This assumes that amount staked by all the predictoors does not change with time, and the rewards remain constant across time as well. (Making such an assumption makes modeling more tractable, yet still useful.)

To further simplify the model, we can assume that each of the n predictoors stakes the same amount sₚ. Then:

The return is then:

From the equation above, we see that returns are positive as long as the numerator is greater than sₚ:

Given a particular value of accuracy Acc, under the conditions presented (all predictoors stake the same amount, stakes and rewards are constant) we obtain the following relationship.

Stake Bounding Formula:

This suggests that for a specific level of accuracy, the amount staked should not exceed the figures on the right side of the equation to ensure a positive return, according to the assumptions made.

4. Stake Amount Guidelines

Here are some general guidelines on how much to stake.

• The lower-bound sales for a given week are the Data Farming Rewards in OCEAN and in ROSE, on the order of 37,500 OCEAN and 20,000ROSE. You can use this for r_f in the formulae.
• Higher accuracy levels lessen the likelihood of negative returns.
• When determining the stake amount, consider the stakes of other users. The system allocates rewards based not only on accuracy but also on the stake amount. Hence, in each epoch, users who predict correctly and have a larger stake will receive a larger share of the rewards.
• Start with very small stakes, where you definitely aren’t bounded by sales. Get a feel for how everything works end-to-end, and how the variables above relate in practice. Only when you feel more confident should you increase

Finally — hopefully to state the obvious — it’s unwise to stake all your OCEAN into a single prediction even if your model’s accuracy is extremely high. Because if you’re wrong then you lose it all! All the models above assume that amount staked is significantly lower than this. However there is a way to bound this amount too, such that you never risk it all yet maximize profit: Kelly Criterion. It’s the original “right-size” approach to making bets in poker and trading, and applies equally here.

4. On Formula Accuracy

The two scenarios presented above are relatively simple due to making specific assumptions. Nonetheless, we hope they prove useful in practice.

For predictoors who want to further maximize their return by improving formula accuracy further, here are some possibilities (at the cost of higher formula complexity).

• Fixed predictoors’ stake → stake changing over time → real-time estimates of stake
• Fixed prediction model accuracy → real-time estimates of accuracy
• Rewards lower bound from Data Farming → add organic sales estimate → real-time estimates of organic sales
• Ignore gas fees → add gas fees estimate → real-time estimate of gas fees
• (and more)

5. Conclusion

This article is for predictoors — people running prediction bots in Ocean Predictoor ecosystem.

If you stake more, you can earn more, but only to a point. Earnings are bounded by sales of the feeds. So if you stake too much, you will lose $.

This article provided equations that bound how much to stake, along with general tips.

Final Note

⚠️ You will lose money as a predictoor if your $ out exceeds your $ in. If you have low accuracy you’ll have your stake slashed a lot. Do account for gas fees, compute costs, and more. Everything you do is your responsibility, at your discretion. None of this blog is financial advice.

About Predictoor & Ocean Protocol

In Ocean Predictoor, people run AI-powered prediction bots or trading bots on crypto price feeds to earn $. Follow Predictoor on Twitter, and get support in discord. Track progress on GitHub at pdr-backend and more. Predictoor runs on Oasis Sapphire confidential EVM chain.

Predictoor is powered by Ocean Protocol, which provides tools to privately & securely publish, exchange, and consume data. Follow Ocean on Twitter or Telegram.