# Synth Subnet - Miner Performance Review #14

**Published by:** [Synth](https://paragraph.com/@synthdata/)
**Published on:** 2025-06-17
**URL:** https://paragraph.com/@synthdata/synth-subnet-miner-performance-review-14

## Content

Data From 2025-06-09 To 2025-06-15BTC and ETH VolatilityFigure 1 shows the hourly log returns and volatility observed throughout the week. Higher ETH volatility and the BTC/ETH correlation are the most notable patterns in the two graphs.Miners PerformanceThis week’s analysis focuses on four miners: 32, 170, 200, and 234. Two of these miners were in the top 3 both at the beginning and end of the week. Miner 170 climbed into the top rankings during the week—ultimately finishing first—while Miner 32, initially in the top 3, dropped out by the end. As usual, we evaluate performance using two key metrics:CRPS (Continuous Ranked Probability Score) for BTC and ETH forecastsLeaderboard ScoresContinuous Ranked Probability Score (CRPS). Lower CRPS values indicate better predictive accuracy. As observed last week, CRPS values for both BTC and ETH were very close among the top miners, indicating a tightening competition. This suggests that success increasingly depends on accurately modeling price nuances.Leaderboard ScoresThese are exponentially weighted averages of past CRPS values over a 10-day window. Lower scores indicate better performance. The leaderboard continues to reflect close competition among top miners. Miner 170 entered the top 3 on Friday, June 13th, and reached the top position over the weekend.Miners’ StrategiesThis section analyzes summary statistics—volatility, kurtosis, and skewness—to detect potential strategic differences among top miners for both BTC and ETH. Summary Statistic Definitions:Volatility: Variability in forecasted returnsKurtosis: Measures the "tailedness" of the distribution—how likely extreme events areSkewness: Measures the asymmetry of return distributions (0 = symmetric)Figure 4 presents these statistics across the week for the four miners.Modeling strategies appear broadly similar across top miners, with only slight differences—particularly in how Miners 234 and 200 modeled BTC kurtosis compared to the others. Miners’ Ranking vs. Summary Statistics: All miners who submitted forecasts were ranked by total weekly rewards. Figure 5 shows how summary statistics vary across different ranking tiers.As seen last week, lower-ranked miners struggled to match top-tier performance, especially in volatility and kurtosis for both BTC and ETH. Still, it's encouraging that some lower-tier miners attempted to model return distribution kurtosis, even if results fell short. Such efforts are likely to be rewarded over time as model quality improves.Analysis by AxonsTo assess performance at the group level, we analyzed:Axon size (number of miners)Average and total rewards per axonAxon-level summary statistics (BTC forecasts)Figure 6 shows the number of miners per axon at the end of the week.As in the previous week, axons 116.202.53.142 and 95.216.99.113 had the largest memberships, with 49 and 33 miners, respectively. Figure 7 shows the evolution of average (per miner) and total daily rewards for the top 10 axons by average rewards.Axons 35.77.6.189, 3.112.97.164, 35.78.218.102, and 18.183.47.137 achieved the highest average and total daily rewards. Notably, the two largest axons were not among the top earners—suggesting that a larger group size does not guarantee greater rewards. Furthermore, the reward system's updated parameters now encourage forecast diversification. Miners from the same axon who submit identical path sets are penalized, as identical paths receive identical scores—and poor-performing models will impact all such miners equally. Figure 8 compares the top 10 axons (by average rewards) based on their BTC modeling characteristics: intraday average volatility, its variability, and average intraday kurtosis.There is a clear difference in how the top four axons model volatility. These axons tend to produce forecasts with greater and more adaptive volatility. They also place more emphasis on modeling kurtosis. In contrast, lower-ranked axons often produce less volatile forecasts and seem to neglect kurtosis modeling. One exception was axon 186.233.184.223, which made some effort in kurtosis modeling but did not adapt forecast volatility as much as the top axons.TakeawaysThis week continued the trend of tightening competition. Top-ranking miners submitted high-quality forecasts with very close scores. A sharper distinction emerged between high-reward axons and others, largely due to superior modeling strategies among the top axons. While the current market and volatility regime appears to favor these top-performing axons, lower-tier axons and miners are encouraged to keep improving their models. Leaderboard shifts are not uncommon, as seen in previous weeks. Finally, the recent change in EMA calculation parameters seems to have had two effects:Large axons do not necessarily yield larger total rewards unless their models are competitive.Lack of diversification within axons is penalized. Submitting identical paths results in shared outcomes, and when the model underperforms, all miners using it suffer lower rewards.

## Publication Information

- [Synth](https://paragraph.com/@synthdata/): Publication homepage
- [All Posts](https://paragraph.com/@synthdata/): More posts from this publication
- [RSS Feed](https://api.paragraph.com/blogs/rss/@synthdata): Subscribe to updates