Purpose

Recent studies have shown that low-volatility shares outperform high-volatility shares. Given the conventional finance theory that risk drives return, this study aims to investigate and attempt to explain the presence of the low-risk anomaly (LRA) in South Africa.

Design/methodology/approach

Using share prices from 1990 to 2016, various buy-and-hold strategies are constructed to determine the return to an investor attempting to capitalise on such an anomaly. These strategies involve combinations relating to a price filter, the calculation of risk and volatility, value-weighting or equal-weighting of portfolios and the window period to construct said portfolios.

Findings

It was found that the LRA exists on the Johannesburg Stock Exchange (JSE_=) when using univariate sorts, without controlling for the size or value effect. When using multivariate portfolio sorts (size and volatility or value and volatility), it was found that the LRA does not exist on the JSE under the majority of risk proxies, but particularly prevalent when downside risk is used. This loosely points towards a potential “inverse momentum” effect where low-return portfolios outperform their counterparts.

Originality/value

In general, it is established that the risk–return relationship is non-linear and deterministic under traditional proxies, but improves to being somewhat, but not completely, linear under a Kalman filter. The Kalman filter, which can be considered a proxy for learning, does not remove the anomaly in its entirety, indicating that behavioural approaches are needed to explain such phenomena.

中文翻译：

---

调查南非的低风险异常

目的

最近的研究表明，低波动性股票的表现优于高波动性股票。鉴于风险驱动回报的传统金融理论，本研究旨在调查并试图解释南非低风险异常（LRA）的存在。

设计/方法/方法

使用 1990 年至 2016 年的股价，构建了各种买入并持有策略，以确定试图利用这种异常情况的投资者的回报。这些策略涉及与价格过滤器、风险和波动性计算、投资组合的价值加权或等权重以及构建所述投资组合的窗口期相关的组合。

发现

当使用单变量排序时，发现 LRA 在约翰内斯堡证券交易所 (JSE_=) 上存在，而没有控制大小或价值效应。当使用多元投资组合分类（规模和波动率或价值和波动率）时，发现在大多数风险代理下 JSE 上不存在 LRA，但在使用下行风险时尤其普遍。这大致指向了潜在的“逆动量”效应，即低回报投资组合的表现优于同类投资组合。

原创性/价值

一般来说，风险-回报关系在传统代理下是非线性和确定性的，但在卡尔曼滤波器下会提高到一定程度但不完全是线性的。卡尔曼滤波器可以被认为是学习的代理，它并没有完全消除异常，表明需要行为方法来解释这种现象。