Hunting Crypto Trading Bots Using Volume Seasonality

Summary

本视频深入探讨了加密货币市场中成交量的季节性变化，重点分析了一周中不同日子和一天中不同时间的成交量行为。利用2018年1月1日至2022年10月的币安数据，演示了如何通过对原始成交量数据进行规范化处理来揭示其季节性特征。分析发现，工作日的成交量普遍高于周末，特别是UTC时间的12点到16点之间，市场活跃度最高。此外，每个常见时间框架的第一分钟成交量较高，这可能揭示了算法交易者的活动模式。视频最后提出了对这些发现的理论解释，并邀请观众提供反馈和讨论。

Takeaways

• 😀 The video analyzes cryptocurrency trading volume seasonality - how volume changes based on day of week and time of day
• 😎 Weekends have lower trading volume than weekdays, showing less activity from professional/institutional traders
• 🚀 Peak trading hours are roughly 12:00 - 16:00 UTC across different cryptocurrency pairs
• ⏰ The first minute of each hour/15 min/5 min period tends to have a spike in trading volume
• 🤔 This spike may be due to trading algorithms placing orders at bar close, which then execute in the next bar
• 😯 By analyzing volume spikes, you can detect consensus opinion of algorithmic traders
• 📈 The raw trading volume data needs to be normalized before analyzing seasonality
• 😴 A 30-day median is used to normalize the daily data
• 💤 A 168-hour (1-week) median is used to normalize the hourly data
• 🥱 A 10080-minute (1-week) median is used to normalize the minute-level data

Q & A

加密货币市场的成交量在一周中的哪些日子最低？

在周末，即星期六和星期日，加密货币市场的成交量显著下降，低于工作日。

如何对原始成交量数据进行规范化处理以揭示季节性变化？

通过取原始成交量与其30天滚动中位数的比值，然后对该比值取自然对数来规范化成交量数据。

为什么要对成交量数据进行规范化处理？

因为原始未经处理的成交量数据不是稳定的，其行为随时间变化不一致，通过规范化处理可以使数据趋于稳定，便于分析。

加密货币市场在一天中的哪些小时成交量最高？

从UTC时间的中午12点到下午4点（12:00-16:00）成交量最高，这一时段市场最为活跃。

周末与工作日的成交量有何不同？

周末的成交量较低，表明周末的市场活动减少，这可能反映了机构活动在周末减少的趋势。

每小时成交量的规范化处理是如何进行的？

通过使用168小时（一周）的滚动中位数来规范化每小时的成交量，以消除周末效应的影响。

在分钟级别的数据分析中，哪些时间点的成交量最高？

在每小时的第一分钟成交量最高，此外，每15分钟和每5分钟的开始也会出现较高的成交量。

高频交易算法在市场成交量中扮演什么角色？

高频交易算法在常见时间框架的第一分钟产生较高成交量，这表明大部分成交量可能来自于算法交易。

为什么在常见时间框架的第一分钟会看到成交量的高峰？

这是因为交易系统通常会在每个时间段结束时分析数据，并在下一个时间段开始时立即下达市场订单，导致成交量激增。

如何利用第一分钟的成交量高峰来理解市场动态？

如果第一分钟的成交量有明显方向，可以作为判断交易算法当前共识意见的一个指标。

Outlines

00:00

📊 加密货币市场的交易量季节性分析

本段落探讨了加密货币市场交易量的日常和小时变化规律。使用2018年1月1日至2022年10月的币安数据，作者发现交易量在不同的日子和时间有显著差异。为了分析季节性变化，必须首先将原始交易量数据标准化，以消除非静态趋势。通过采用30天和168小时的滚动中位数来标准化日和小时数据，使得交易量数据趋于稳定。分析结果显示，周末的交易量显著低于工作日，且在UTC时间的12点到16点间，市场交易量最活跃。此外，作者还研究了不同加密货币对和不同交易所之间的交易量季节性，发现类似的趋势。

05:00

🕒 交易量的时间段分析与算法交易的影响

本段讨论了交易量的三个主要趋势：工作日比周末交易量高、一天中有特定的高峰小时和一些时间框架开始时的第一分钟内交易量上升。这些趋势揭示了机构活动的痕迹和市场的全球性特征。作者提出了一个理论，认为交易系统在每个时间框架结束时做出交易决策，导致新周期开始时的交易量激增。如果第一分钟内有明显的交易量增加，这可能表明算法交易在起作用。最后，作者鼓励观众讨论并为进一步的研究提供意见。

Keywords

💡volume

The amount of cryptocurrency traded during a particular time period. The analysis examines volume patterns over different time frames like daily, hourly, and by minute.

💡seasonality

Recurring and predictable patterns in volume over different time periods. The analysis aims to uncover volume seasonality in crypto markets.

💡normalize

Transform the raw volume data to make it stationary for consistent comparison across time. This is done by dividing by a rolling median and taking the logarithm.

💡peak hours

Times of day when trading volume tends to be highest, typically 12-16 UTC across different coins and exchanges.

💡weekend effect

Volume tends to be lower on weekends than weekdays, indicating less activity from institutional traders.

💡first minute volume

Volume in the first minute of each time frame (day, hour, 15 min, 5 min) tends to spike. This suggests trading algorithms placing orders at bar closes.

💡trading algorithms

Automated trading systems that analyze market data and place orders. The first minute volume spikes indicate their activity.

💡stationary

A statistical property meaning the distribution has a consistent mean and variance over time. Needed for comparing averages.

💡UTC

Coordinated Universal Time, the standard for tracking time used in global financial markets. Replaces references to time zones.

💡consensus opinion

If trading algorithms move price strongly in the first minute, it indicates their consensus view on market direction.

Highlights

标准化加密市场中的交易量数据以分析季节性模式

星期六和星期日的交易量明显偏低,工作日的交易量相对较高

每天中午12点至下午4点(UTC时间)是交易量的高峰时段

每天的第一分钟交易量明显偏高,表明算法交易的脚印

常见时间框架(一天、一小时、15分钟)的第一分钟交易量偏高

交易算法通常在某个时间框架结束时决定其行动,导致下一分钟交易量增加

如果第一分钟价格变动很大,可以推断出该时段算法交易的共识意见

周末效应表明周六日交易量低于工作日,反映了机构活动的印记

加密市场每天高峰小时对应不同地区的交易高峰时段

交易量高峰时段和第一分钟交易量偏高是整个市场的普遍规律

进一步研究周末和工作日的价格行为差异可能产生有趣发现

标准化方法使原始交易量数据更加平稳,便于분析其季节性模式

使用移动中位数和对数转换将原始交易量变换为可分析的形式

变换后的交易量分布更趋于正常,方差波动更小,便于提取其规律

分离 weekends 和 weekdays 的交易量曲线,可以清楚看到两者差异

Transcripts

today we're going to be talking about

volume seasonality we're going to be

looking at how volume behaves in the

cryptocurrency markets based on the day

of the week and the time of the day

you'll learn some details about how

volume behaves in cryptocurrency markets

and I'll show you a way to see what

algorithmic Traders might be up to

we'll be using data from binance between

January 1st 2018 and October 2022. to

uncover how volume tends to behave at

different times we will average the

volume across all instances of a common

time such as Mondays or Tuesdays for

daily data or 11 A.M and 2 p.m for

hourly data however before we can do

this we need to normalize the volume

data as raw unaltered volume is not

stationary its behavior is not

consistent across time here we have the

raw volume for Bitcoin tether over our

test period by looking we can observe

that the raw volume exhibits many

shorter term Trends and doesn't have a

consistent level at hovers around put

differently it does not have a

stationary mean value to better

illustrate this I took the average

volume of each year in the test set and

plotted it

since we're seeking to average volume

readings at different times to extract

its seasonal Behavior we have to

transform the raw volume to induce some

stationarity

to do this I take a rolling median of

the volume for daily data I chose a

rolling period of 30 days this decision

is arbitrary but 30 days about a month

is reasonable I divide the raw volume by

The Rolling median and take the natural

logarithm of that quotient

once this is done here's the result the

normalized volume we can see that the

transform series tends to hover around

zero and is fairly consistent throughout

time now if we again take the average of

each year they are very similar looking

at the histogram of our transform series

we can see that the distribution has a

nice Bell shape I should mention that

this new series is not perfectly

stationary in a statistical sense but

it's good enough for our purposes

this is our normalized volume average by

the day of the week the main feature

here is that the weekends Saturday and

Sunday have a significant drop off a

consistently lower level and the

weekdays have a consistent level but are

much higher than the weekends

now we're going to look at how volume

behaves given the time of the day all of

the times I'll give will be in UTC

coordinated universal time also known as

GMT or Zulu time military time here's

the full view of our normalized volume

series for the hourly time series for

this I used a rolling median period of

168 hours which is one week that is to

cancel out the weekend effect that we

just saw in the daily analysis the

hourly normalized still has that nice

bell-shaped distribution and here is the

average normalized volume for each hour

of the day there's a couple notable

things here first there's some peak

hours roughly from hour 12 to hour 16

noon to four o'clock in UTC and a little

bit of wider range for peak hours would

be from hour a to hour 17. this section

of the day is when the market tends to

be the most active there is the most

volume going through it another

interesting feature is the first hour of

the day has a spike this is important

and we'll see more of that soon because

we just saw that weekends to never lower

volume than weekdays I did the same

hourly study but separated weekdays from

weekends to see if the peak hours differ

and when they're plotted against each

other we can see that the same peak

hours hold throughout the weekends but

the weekend levels are of course slower

to visualize this differently here's a

heat map mapping every hour of the day

with every weekday I was curious to see

if the peak hours held up throughout the

entire data set so I redid the same

seasonality averaging but for each year

in the data set and as you can see 2018

through 2022 that same shape holds each

year has that first hour Spike and the

peak hours are pretty much the same I

also did the same thing with coup coins

Bitcoin tether pair to see if the peak

hours held across exchanges and as you

can see they do they roughly have the

same shape and also did the same

seasonality averaging for a few

different common cryptocurrency Pairs

and as you can see they all have roughly

the same peak hours and same kind of

shape so these peak hours appear to be a

market-wide phenomenon not just for

Bitcoin now we're going to look at

minute data the rolling medium period I

used for minute data was one week in

minutes 24 times 60 times seven I don't

know what that is you can do the math so

here's our normalized volume for each

minute of an hour and as you can see

that on average the first minute has the

most volume compared to any other minute

and if we look a little closer we can

see there's a slightly less prominent

Spike at each 15 minute interval and if

we look even closer we can see that

there's a smaller Spike at each five

minute interval

here's the volume average for each

minute of the day there's

1440 minutes in a day we can see that

beginning of the day Spike we saw in the

hourly data and look a little closer we

can see a large amount of activity that

takes place in the first few minutes of

the day here is the volume for each

minute of the day with each hour marks

as you can see that the beginning of

each hour has a spike and relatively

higher volume compared to neighboring

minutes

foreign

thing but with each 15 minutes marked

and as you can see at the beginning of

each 15 minute period the volume tends

to be higher than neighboring minutes

and it's fairly consistent throughout

the day

we have found three Tendencies of volume

one there's more volume on weekdays than

weekends the weekend effect two there

are peak hours of the day where the

market has more volume and three the

first minute of common time frames tends

to have higher volume let's talk about

each of these the weekend effect shows a

tendency for lower volume on the

weekdays I believe this difference shows

the footprint of institutional activity

professional entities would reasonably

take the weekends off I think further

study of price Behavior Beyond volume on

weekends versus weekdays May yield

interesting results perhaps we'll make

another video looking into this comment

if you're interested the cryptocurrency

market tends to have higher volume

between 12 and 16 UTC this suggests more

Traders tend to be awake and active

during these times roughly speaking

these hours tend to fall in the morning

for the U.S the afternoon for Europe and

the evening for Asia at least for volume

the market tends to behave differently

during these times in my opinion our

most interesting binding is the higher

average volume in the first minute of

common time frames such as one day one

hour 15 minutes and 5 minutes I have a

theory about why we see this I've been

developing trading systems for a number

of years and I know that it is common

practice for trading systems to decide

their action at the close of each bar

after the bar is closed a trading

algorithm does whatever analysis it does

and it might place an order if this

order is a market order it will execute

right away and the volume associated

with that order will add to the next

minutes volume this would explain the

high volume we see in the first minute

of common time frames and this Theory

seems reasonable to me and frankly I'm

convinced it's true if you have an

alternative Theory to explain the high

volume in the first minute please leave

it in the comments below I'd like to

hear it so assuming this theory is true

if you look at the first minute of a

common time frame and see a spike in

volume you know that it's likely a large

percentage of that volume is from

Trading algorithms and if that first

minutes bar had a strong Direction you

can get an idea of the consensus opinion

of the trading algorithms at that

current time

alright that's all I have for volume

seasonality thanks for watching

subscribe hit the like button comment

bye

[Music]