通过几个假设（特别是关于method1和df_new），您当前的代码运行在...

%%timeit
# idk what `method1` is so set it to `True`
method1 = True
# idk what `df_new` is so making an 
# empty dataframe to replicate your code
df_new = pd.DataFrame()

data['daily_change'] = data.groupby('title',group_keys=False)['return'].pct_change()
for title in data['title'].unique(): # Iterate through each title
     
    temp_df = data[data['title'] == title].tail(252) # Select the data for a specific title
    if len(temp_df) < 252:
        print(f"{title} has less than 1 year of data, ignore\n")
        continue

    sections = [temp_df.iloc[i:i+63] for i in range(0, 252, 63)] # Divide the data into 4 sections

    if method1:
        result = sum([(section['return'].iloc[-1] / section['return'].iloc[0]) * weight for section, weight in zip(sections, [0.2]*3 + [0.4])]) # Calculate the weighted return
    else:
        # Calculate the weighted return using the daily changes
        result = sum([(1 + section['daily_change']).prod() * weight for section, weight in zip(sections, [0.2]*3 + [0.4])]) - 1

    df_new = pd.concat([df_new, pd.DataFrame({'title': [title], 'result': [result]})], ignore_index=True)

7.48 ms ± 532 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

一些（可能过度优化）改进

import numpy as np
import pandas as pd

# define a function that can be applied over groups
def func(data: pd.DataFrame, method1: bool = True) -> float:
    # take the last 252 rows
    df = data.tail(252)
    # return nothing if less than 252 rows
    if len(df) < 252:
        return
    # storing `weights` at (4,) array
    weights = np.array([*[0.2]*3, 0.4])
    # idk what method1 is so assuming it's a boolean
    if method1:
        # split `df["return"]` into quarters
        sections = np.array(np.array_split(
            ary=df["return"].to_list(), indices_or_sections=4
        ))
        # shave off a few ms...
        result = np.multiply(
            np.divide(sections[:, -1], sections[:, 0]), weights
        ).sum()
    else:
        # split `df["daily_change"]` into quarters
        sections = np.array(np.array_split(
            ary=df["daily_change"].to_list(), indices_or_sections=4
        ))
        # shave off a few ms...
        result = np.multiply(
            np.add(sections, 1).prod(axis=1), weights
        ).sum() - 1
    
    return result

%%timeit
# `method1` is True
data.groupby("title").apply(func, method1=True)

3.68 ms ± 222 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

%%timeit
# `method1` is False
data.groupby("title").apply(func, method1=False)

3.87 ms ± 227 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

根据给定的数据，我能够将时间缩短大约50%。

注解

如果"title"组没有足够的行，函数将不返回任何内容，结果可能如下所示：

dfc = data.sample(1_500, random_state=0)

# these numbers will not be correct due to sampling
print(dfc.groupby("title").apply(func))

title
A         NaN
B         NaN
C         NaN
D         NaN
E    1.282753
F    0.928689
dtype: float64

import numpy as np

for title in data['title'].unique(): # Iterate through each title
     
     temp_df = data[data['title'] == title].tail(252) # Select the data for a specific title
     if len(temp_df) < 252:
         print(f"{title} has less than 1 year of data, ignore\n")
         continue
     
     sections = [temp_df.iloc[i:i+63] for i in range(0, 252, 63)] # Divide the data into 4 sections
     sections_returns = np.array([section['return'] for section in sections])
     sections_daily_change = np.array([section['daily_change'] for section in sections])
     weights = np.array([0.2]*3 + [0.4])
     
     if method1:
         result = np.sum((sections_returns[:, -1] / sections_returns[:, 0]) * weights) # Calculate the weighted return
     else:
         # Calculate the weighted return using the daily changes
         result = np.sum((1 + sections_daily_change).prod(axis=1) * weights) - 1
         
     df_new = pd.concat([df_new, pd.DataFrame({'title': [title], 'result': [result]})], ignore_index=True)

主要变化如下：
使用np.array存储每个部分的收益和每日变化数据使用np.sum计算收益或每日变化的加权和使用prod方法中的axis=1计算沿着行轴的每日变化的乘积这应该会加快大型数据集的计算时间。